Description of the condition
Pneumonia is an infection of the lungs and can be caused by bacteria, viruses, Rickettsia, fungi or parasites. Nearly 100 species have been identified as aetiological agents (Mandell 2010; File 2003; Ruuskanen 2011). Although the pathological definition of pneumonia is clear, the clinical diagnosis is sometimes ambiguous (Appendix 1).
The risk of pneumonia is increased in young children and the elderly. In low-income countries, pneumonia causes two million deaths annually among children under five years of age (Mizgerd 2006; Paynter 2010; Walker 2013). In the USA, pneumonia is the sixth most common cause of death and the most common cause of infection-related deaths (Mandell 2010).
Description of the intervention
Vitamin C was identified in the 1930s as consequence of the search for a substance, the deficiency of which causes scurvy. After its identification there was much interest in its effects on diseases unrelated to scurvy but this role remains unclear.
According to systematic reviews, over two dozen controlled trials have shown that vitamin C shortens the duration of colds (Hemilä 2003; Hemilä 2013a). Five trials have found that vitamin C halved the incidence of colds in participants who endured heavy acute physical stress (Hemilä 1996; Hemilä 2013a), three trials have found that vitamin C halved the exercise-induced decline in forced expiratory volume in one second (FEV
In randomised trials with critically ill intensive care unit (ICU) patients, vitamin C (Tanaka 2000) and the combination of vitamins C and E (Nathens 2002) significantly decreased the length of mechanical ventilation and in a case-control study ICU patients administered vitamins C and E and selenium had shorter ICU stays and lower mortality rates (Collier 2008). Furthermore, dietary vitamin C intake modified the effect of vitamin E on the mortality of older male smokers and this modification was not explained by other substances in fruit and vegetables (Hemilä 2009a).
Although such findings indicate that the effects of vitamin C are not limited to preventing overt scurvy, their practical significance is not yet clear. Two large trials with US male physicians and female health professionals found no benefits of 0.5 G/day of vitamin C (Cook 2007; Sesso 2008) and ≥ 1 G/day of vitamin C had no effect on common cold incidence in the general population (Hemilä 2013a), but these findings are not discordant with the possibility that vitamin C administration might influence, for example, susceptibility to pneumonia or the severity of pneumonia in special conditions.
There is much evidence that vitamin C influences the immune system but its effects may be apparent only in particular conditions. For example, it is possible that variation in vitamin C intake does not influence the immune system in the ordinary Western population because of their relatively high dietary intake levels. Nevertheless, vitamin C might be a limiting factor in populations with low intakes. An extreme example is the high prevalence of frank vitamin C deficiency, apparent as scurvy, in refugee camps in the Horn of Africa: reported to have been up to 44% (WHO 1999b).
In the early 1900s, Alfred Hess carried out extensive studies of scurvy and summarised a large series of autopsy findings: "pneumonia, lobular or lobar, is one of the most frequent complications [of scurvy] and causes of death" and "secondary pneumonias, usually broncho-pneumonic in type, are of common occurrence and in many [scurvy] epidemics constitute the prevailing cause of death" (Hess 1920). He further commented, in a major medical journal a decade later, that in "infantile scurvy . . . a lack of the antiscorbutic factor [vitamin C] which leads to scurvy, at the same time predisposes to infections [particularly of the respiratory tract] . . . Similar susceptibility to infections goes hand in hand with adult scurvy" (Hess 1932). In the early 1900s, Casimir Funk, who coined the term 'vitamin', pointed out that an epidemic of pneumonia in the Sudan disappeared when antiscorbutic (vitamin C-containing) treatment was given to the numerous cases of scurvy which appeared at about the same time (Robertson 1934).
Vitamin C deficiency is not just of historical relevance. Cases of scurvy in hospitals have been described in several recent case reports, for example Holley 2011 and Smith 2011, and one survey estimated that about 10% of hospitalised elderly patients had scurvy (Raynaud-Simon 2010). In the UK, 25% of men and 16% of women from low-income populations had vitamin C deficiency (< 11 μmol/L) (Mosdøl 2008) and in the USA 7% of healthy middle-class participants had vitamin C deficiency (Schleicher 2009). Thus, if pneumonia risk is increased by low intakes of vitamin C, this issue might also be important in parts of high-income countries.
Furthermore, vitamin C metabolism is affected by various infections, including pneumonia, as indicated by decreased levels in plasma, leucocytes and urine (Bakaev 2004; Bhoite 2011; Cemek 2006; Chakrabarti 1955; Hemilä 1997a; Hemilä 1999b; Hemilä 2006; Hunt 1994; Mochalkin 1970). These changes in metabolism mean that vitamin C might have a treatment effect on pneumonia patients irrespective of their dietary intake. Since the 1930s, a few German and US physicians have proposed that vitamin C might be beneficial in the treatment of pneumonia (Bohnholtzer 1937; Hochwald 1937). Gander and Niederberger concluded from a series of 15 cases that "the general condition is always favourably influenced [by vitamin C] to a noticeable extent, as is the convalescence, which proceeds better and more quickly than in cases of pneumonia which are not treated with vitamin C" (Gander 1936). Benefit from intravenous vitamin C was reported in a series of over 40 cases (Klenner 1948; Klenner 1951) and in three cases of viral pneumonia (Dalton 1962). Large oral doses of vitamin C were also claimed to be beneficial in patients with viral pneumonia (Cathcart 1981; Luberoff 1978).
The effect of vitamin C on the common cold has been studied extensively. A major finding from the trials is the heterogeneity in its effects. Although the largest trials found no effect on common cold incidence, it was significantly reduced in trials with participants under heavy acute physical stress and in British males, the latter being explained as the result of a diet low in vitamin C when the studies were carried out (Hemilä 1996; Hemilä 1997b; Hemilä 2006; Hemilä 2013a). Thus, the effects of vitamin C on other respiratory infections, such as pneumonia, might also be modified by various factors, such as physical stress and dietary vitamin C intake. Furthermore, two large trials found considerable divergence in the effects of vitamin C depending on the type of cold. Vitamin C decreased the incidence of 'chest colds' (-18%; cough or other chest symptoms) but not of 'simple colds' (+1%; runny nose or sneezing) (Elwood 1976; Hemilä 1997b). Similarly, vitamin C decreased the incidence of 'throat colds' (-21%) but not 'nose colds' (-2%) (Anderson 1973; Hemilä 1997b). These two trials thus suggest that vitamin C might have a greater effect on infections affecting the lower respiratory tract.
In close parallel with vitamin C, lipid-soluble vitamin E is interesting as these two antioxidants interact; vitamin C reduces oxidised vitamin E back to the reduced form (Hamilton 2000; Hemilä 2006; Hemilä 2009a; Packer 1979). In a large-scale trial the effect of vitamin E on the risk of pneumonia was significantly modified simultaneously by smoking and leisure-time physical activity, indicating heterogeneity in vitamin E effects between population groups (Hemilä 2011a). Even though direct extrapolation of findings from vitamin E studies to vitamin C are unjustified, the notion that various factors may modify the effects of antioxidants is fundamentally important in restricting broad generalisations from individual trials, irrespective of whether the finding is positive or negative and whether or not the trial is large and carefully conducted.
Approximately 10 mg/day of vitamin C prevents scurvy but the safe dose range extends to grams per day. In the US nutritional recommendations, the 'tolerable upper intake level' is stated to be 2 G/day for adults. However, the basis for this upper limit is the appearance of diarrhea (IOM 2000) which is a trivial adverse effect that disappears quickly with a reduction in intake. Furthermore, it has been stated that patients with pneumonia can take 100 G/day of vitamin C without developing diarrhea, possibly because of the changes in vitamin C metabolism (Cathcart 1981).
How the intervention might work
Vitamin C is an antioxidant and its effects may be most pronounced under conditions when oxidative stress is increased. Viral and bacterial infections lead to the activation of leukocytes, which generate reactive oxygen and nitrogen species that oxidise extracellular vitamin C (Akaike 2001; Galley 1996; Hemilä 1984). This mechanism explains the decrease in vitamin C levels in pneumonia patients and in other patients with infections (see above). The oxidised form of vitamin C is imported to leukocytes where it is reduced back to vitamin C and its concentration can become very high (Nualart 2003; Wang 1997). Exercise also causes oxidative stress because of increased oxygen consumption (Ashton 1999; Powers 2008).
In the immune system, the major role of vitamin C seems to be as a physiological antioxidant, protecting host cells against oxidative stress caused by infections. In various experimental settings vitamin C has, for example, affected random migration and chemotaxis of phagocytes (Goetzl 1974), the transformation of influenza virus-infected lymphocytes (Manzella 1979), the production of interferon (Siegel 1975), the replication of viruses (Bissell 1980) and gene expression of monocyte adhesion molecules (Rayment 2003); see reviews in: Beisel 1982, Hemilä 1997a, Thomas 1978 and Webb 2007.
In dozens of animal studies, vitamin C increased resistance against diverse viral and bacterial infections and against purified bacterial toxins (Hemilä 2006). In a laboratory study, vitamin C deficiency increased the incidence of pneumonia in rhesus monkeys (Sabin 1939). In mice, influenza infection decreased vitamin C concentration in bronchoalveolar lavage fluid (Buffinton 1992) and, also in mice, vitamin C deficiency increased lung pathology caused by influenza infection (Li 2006).
See the Discussion for further considerations on the possible mechanisms of vitamin C effect on pneumonia.
Why it is important to do this review
Pneumonia is a fairly common and severe infection and vitamin C is a safe and inexpensive essential nutrient. The possibility that vitamin C might affect susceptibility to pneumonia, even in restricted population groups, is worthy of examination. Similarly, the possibility that vitamin C treatment might affect the duration or severity of pneumonia, or both, is worthy of systematic consideration. One previous meta-analysis assessed the preventive effects of vitamin C on pneumonia (Hemilä 1997c) but the therapeutic effect on pneumonia was not assessed systematically before this review.
Links to the publications cited in this section, for which full-text versions are available, can be found at http://www.mv.helsinki.fi/home/hemila/CP.
To assess the prophylactic and therapeutic effects of vitamin C on pneumonia.
Criteria for considering studies for this review
Types of studies
For preventive trials of vitamin C supplementation we used controlled trials. The use of placebo was not required as it seems unlikely that being aware or not of taking vitamin C would affect the occurrence of a severe infection. Also, since a recent meta-analysis of trials comparing a placebo group with a no treatment group found no evidence of a placebo effect on binary outcomes (Hrobjartsson 2001; Hrobjartsson 2010), there is no empirical evidence indicating that the placebo effect might affect the occurrence of pneumonia.
For treatment trials of vitamin C on the severity and duration of pneumonia we used placebo-controlled trials since the outcome (for example, severity) may be affected by the awareness of the treatment by the patients. Also, the recent meta-analysis of trials comparing a placebo group with a no treatment group found evidence of a placebo effect in trials focusing on pain (Hrobjartsson 2001; Hrobjartsson 2010). A placebo control may, therefore, be crucial for the validity of treatment observations.
Types of participants
For prevention trials there was no age restriction in the participants.
For treatment trials we restricted trials to participants with pneumonia (both community-acquired and hospital-acquired (nosocomial) pneumonia) with no age restrictions.
Types of interventions
Administration of vitamin C (ascorbic acid or its salts) to one trial group, either orally or intravenously. There were no restrictions on the dosage and frequency of administration of vitamin C and treatment trials with a single dose were also considered. We excluded trials in which vitamin C was administered along with other substances, such as other vitamins. We included studies in which vitamin C had a co-intervention if the control group only had the co-intervention so that the only difference was vitamin C administration.
Types of outcome measures
Community-acquired pneumonia and hospital-acquired (nosocomial) pneumonia are very different types of medical conditions and we analysed them separately.
The preventive effect of vitamin C:
- incidence of pneumonia during vitamin C supplementation.
The treatment effect of vitamin C:
- duration and severity of pneumonic episodes;
- death caused by pneumonia.
Hospital-acquired (nosocomial) pneumonia
The preventive effect of vitamin C:
- incidence of pneumonia during vitamin C supplementation.
The treatment effect of vitamin C:
- duration and severity of pneumonic episodes;
- death caused by pneumonia.
For our review pneumonia was defined operationally as the disease that the original trial authors classified as pneumonia. The basis of the diagnosis by the original authors is described in the Description of studies section. We did not require that the pneumonia diagnosis was based on chest X-radiography (CXR) but we also accepted a clinical diagnosis of pneumonia (see Appendix 1 for detailed comments on this issue).
- Laboratory findings, such as C-reactive protein or erythrocyte sedimentation rate.
- CXR changes and body temperature changes during treatment.
Search methods for identification of studies
For this 2013 update we searched the Cochrane Central Register of Controlled Trials (CENTRAL) 2013, Issue 3, part of The Cochrane Library, www.thecochranelibrary.com (accessed 8 April 2013) which contains the Acute Respiratory Infections Group's Specialised Register, MEDLINE (1950 to March week 4, 2013), EMBASE (1974 to April 2013) and Web of Science (1945 to April 2013). Details of the original search are shown in Appendix 2.
We searched CENTRAL and MEDLINE using the following search strategy. We did not use a filter to identify randomised trials as there were few results. We adapted the search for EMBASE (Appendix 3) and Web of Science (Appendix 4).
1 exp Pneumonia/
4 exp Bronchitis/
7 exp Ascorbic Acid/
10 vitamin c.tw,nm.
11 vit c.tw,nm.
13 6 and 12
Searching other resources
Previously, Briggs 1984 carried out extensive searches of the literature and published a bibliography containing 413 references to papers related to vitamin C and infections. We perused the Briggs bibliography and other pertinent reviews and publications along with the results of the database searches. There were no language restrictions in the literature searches.
Data collection and analysis
Selection of studies
The contact author (HH) searched the literature and both review authors (HH, PL) independently assessed the extracted titles and abstracts to identify potentially relevant articles. We excluded trials failing to meet the inclusion criteria. When we disagreed on the relevance of an article, we discussed it until we reached a consensus.
Data extraction and management
Both review authors (HH, PL) independently extracted relevant data from the articles selected. When we differed in the interpretation of study findings we sought a consensus. In their report, Hunt 1994 published the respiratory symptom scores for all participants and for the most severely ill patients; for this review we calculated the scores for the less ill patients.
Assessment of risk of bias in included studies
We recorded the following quality features of the trials: allocation concealment, blinding, proportion of drop-outs and other relevant features that may limit the validity of the trial. We did not calculate any quality scores for the selected trials since "quality scores are at best useless and at worst misleading" (Greenland 1994). We agree with the Shapiro 1997 comment that quality is best evaluated qualitatively. TheCochrane Handbook for Systematic Reviews of Interventions also states that "The use of scales for assessing quality or risk of bias is explicitly discouraged in Cochrane reviews" (Higgins 2011).
Measures of treatment effect
We planned to calculate risk ratios (RR) for dichotomous outcome variables. However, in the identified prophylactic trials the number of pneumonia cases in the vitamin C groups was very low (zero to two cases) and, therefore, we decided to use the Peto method for calculating the odds ratio (OR), which does not need corrections for zero cell counts (Higgins 2011).
Also, with only a few cases observed in the prophylactic trial groups, the mid-P value is the most appropriate method to calculate the P values for the differences in the treatment groups (Hemilä 2006; Lydersen 2009) and was used when comparing groups with small numbers of cases. We used two-tailed P values in this review.
Unit of analysis issues
The Glazebrook 1942 study reported the number of pneumonia cases per seven administrative groups of the school. Thus, the unit of analysis is the group of schoolchildren in the administrative division. Glazebrook states that "The youths of one division worked as a unit, and occupied certain tables in the dining hall. To some extent each division occupied particular dormitories but this separation was not absolute and there was a fair amount of mixing of divisions in the sleeping quarters. Sleeping and feeding conditions were, of course, the same for all divisions. Careful records had been kept of the incidence of all infections for 1½ years before the observations described here were begun. In the preceding year there had been an epidemic of tonsillitis, which had affected all the divisions uniformly, so that they could not be regarded as separate units within the larger population." Therefore, we consider that the schoolboys had a similar risk of pneumonia in each division and we carried out our primary analysis by the vitamin C and the no vitamin C groups. However, as a sensitivity analysis, we also analysed the data by administrative units (see Results).
Other studies included in our analyses do not have unit of analysis concerns.
Dealing with missing data
None of the trials had missing data that we needed to impute.
Assessment of heterogeneity
We assessed the heterogeneity of comparisons by using the I
Assessment of reporting biases
We planned that if a number of trials were available with sufficient uniformity in settings and outcome definitions, we would pool the data; but, if the trials were heterogeneous, either statistically or clinically, we would present them separately. There is no statistical heterogeneity in Analysis 1.1 but the studies are clinically so divergent that pooling was inappropriate.
Subgroup analysis and investigation of heterogeneity
We included both community-acquired and hospital-acquired (nosocomial) pneumonia in this review. Due to their substantial clinical differences, we analysed them separately.
We did not require that pneumonia definition was based on CXR (Appendix 1) but we planned to carry out subgroup analyses based on the rigour of outcome definition (CXR or not) and on the level of blinding of outcome assessments. Given the trials we identified, this did not differ from the sensitivity analysis based on the methodological quality of the trials.
We did not set limits on the vitamin C doses for the inclusion of trials but we planned to carry out subgroup analyses based on dosage. In the preventive trials, we set the limit of subgroup analysis to 100 mg/day, since it is close to the dosage leading to maximum vitamin C plasma levels in healthy people. In the treatment trials, we set the limit of subgroup analysis to 1000 mg/day, since there is evidence of changes in vitamin C metabolism in infections and larger doses might be needed for significant therapeutic effects. We did not find suitable variation in the doses for a between-study subgroup analysis but there was within-study variation in the Mochalkin 1970 trial ( Table 1).
Two of the identified trials were double-blind, placebo-controlled, randomised controlled trials (RCTs) (Hunt 1994; Pitt 1979) whereas three studies were methodologically less satisfactory (Glazebrook 1942; Kimbarowski 1967; Mochalkin 1970). We carried out sensitivity analysis by excluding the latter three methodologically poorer quality trials.
Glazebrook 1942 had a unit of observation of an administrative group in a boarding school. In the primary analysis, we assumed a similar risk for each participant in each administrative unit but as a sensitivity analysis we also analysed their data by the administrative groups.
Description of studies
Results of the search
We identified no new trials for inclusion in the 2013 update searches (out of 125 records recovered). The MEDLINE search for the 2013 update retrieved 12 publications, the EMBASE search retrieved 119, the Web of Science search 21 and the CENTRAL search 0 publications. From these search results we found no new trials in addition to the earlier identified three controlled trials which provided data pertinent to the prevention of pneumonia with vitamin C supplementation and two trials which provided data on the therapeutic effect of vitamin C. For earlier searches, see Appendix 2. However, in the 2013 update we identified an earlier study which had been missed since 'pneumonia' was not listed in its key words (Tanaka 2000).
The main features of the included trials are summarised in the Characteristics of included studies table. The methods are described here in more detail, largely using direct excerpts from the original papers as these show the strengths and weaknesses of the trials in the words of the original trial authors.
Only six trials were identified. Their main features are briefly summarised in the table Characteristics of included studies but described in detail in this section because their weaknesses and strengths are crucial when considering the validity of their findings and the possibility of extrapolating their findings.
Tanaka 2000 was primarily interested in resuscitation fluid volume requirements and oedema generation in severely burnt patients. The study is reported as a "randomised study" but the methods section states that "randomisation was performed according to the month of admission" suggesting that randomisation may have been used as a synonym for allocation. There is no description of the level of blinding. Incidence of pneumonia is reported as a secondary outcome but the criteria for diagnosis are not described.
Hunt 1994 stated that "The patients enrolled into this . . . study were suffering from acute bronchitis (often acute exacerbation of chronic bronchitis) or bronchopneumonia. Patients suspected or known to be suffering from lung cancer were excluded from the study, as were those who were judged by the clinician to be at high risk of death within a day or two of admission" (page 213). Thus the patients in this study were a mixture of bronchitis and pneumonia patients, whereas in our methods section our purpose was to focus on pneumonia. However, with the soft clinical definition of pneumonia, as mentioned in the Background and Types of outcome measures sections and the high false negative proportion in CXR versus high-resolution computed tomography (HRCT) comparison (Appendix 1), we included this trial in our analysis. Nevertheless, the wider definition of lower respiratory tract infection in this trial needs to be considered when drawing conclusions. "The patients were enrolled over a period of three years and were admitted mainly in the winter months. . . acute respiratory infection had, in all cases, been the primary reason for hospitalisation" (page 213). "For consistency all clinical assessments were performed by the same Associate Specialist. Three main diagnostic features of infective respiratory conditions, namely cough, breathlessness and radiographic evidence of chest infection were used. Each was scored by the clinician according to severity. . . Then for each person, at each assessment interval, his or her three main diagnostic feature scores were added to give the 'total respiratory score'. By this procedure, the worst score that could be achieved by the most severely ill patient (whilst still alive) was 9, whilst those who were completely well with regard to the respiratory condition would score 3. A score of 10 was given for subjects who died during the trial. . . Assessments were made on admission (0 weeks) and at 2 and 4 weeks after admission. If patients were discharged from hospital as 'well' before 4 weeks, therapy was discontinued and they were assumed to remain well for up to 4 weeks, for the purpose of clinical scoring (none of the patients discharged were readmitted during their 4 week assessment period)" (page 213). "The clinical score results were approximately normally distributed..." (page 214), which allowed us to use the t-test in the comparison of the clinical score values. "After the initial clinical assessment . . . the patients commenced placebo or vitamin C therapy to which they were allocated on a randomised 'double-blind' basis. This was in addition to their normal medication" (page 213). Thus, the test of vitamin C effects was "over and above those of normal medication (mainly antibiotics and cough medicines) to which all participants were exposed" (page 217). "The vitamin C and placebo tablets were indistinguishable from each other by look or taste" (page 213). "None of the subjects who died on the trial had any secondary diagnosis, including ischaemic heart disease, and death was attributed directly to respiratory infection in each case" (page 217). At baseline, the mean plasma vitamin C level was 23.3 μmol/L and 35% of patients had a vitamin C level lower than 11.4 µmol/L (page 215). After four weeks, the vitamin C level was 94.9 μmol/L (+307%) in the vitamin C group but only 24.4 μmol/L (+5%) in the placebo group (page 215).
The authors of Pitt 1979 were primarily interested in the effect of vitamin C on the incidence of the common cold. However, other severe respiratory infections including pneumonia were also recorded. "The participants were male marine recruits who underwent 11 weeks of recruit training at Parris Island, South Carolina in October to December. . . Pill taking did not begin until the recruit's third week at Parris Island" (page 908). "These 862 recruits were assigned randomly to either the vitamin C or placebo group from a list of consecutive numbers randomised in pairs. Randomisation was carried out by individual recruits within each platoon" (page 908). "Of the 862 recruits who began taking the pills, 64 recruits (34, vitamin C; 30, placebo) were removed from their platoons by the US Marine Corps for further training or for discharge during the eight-week study period. An additional 123 recruits (64: vitamin C; 59: placebo) were excluded from the final analysis because they did not continue to take their pills for the eight-week study period. One additional recruit was eliminated from the vitamin C group because of recurrent urticaria related to taking the tablets" (page 909). "Before the initiation of pill taking, each recruit received adenovirus 4 and influenza vaccines and either intramuscular penicillin G benzathine or oral erythromycin estolate as streptococcal prophylaxis" (page 908). "Pill taking was supervised and observed by the drill instructors in each platoon. Neither the recruits or drill instructors nor the physicians and corpsmen who treated the recruits were aware of which pill any individual was taking" . . . "The placebo tablets were formulated from citric acid and were indistinguishable in appearance and taste from the vitamin C tablets" (page 908). "Pneumonia developed in eight recruits. . . Each of these eight recruits had typical roentgenographic and physical signs of pneumonia, although five recruits were febrile and only four recruits had elevated white blood cell counts. Pneumococci were isolated from the sputum in three recruits and seen intracellularly on Gram's stain in two other recruits. Two of these recruits also had four-fold increases in parainfluenza titers . . . Each of these recruits returned to his platoon after a mean Medical Dispensary stay of 4.4 days" (page 910). Pitt and Costrini did not estimate dietary vitamin C intake; however, their participants' mean vitamin C plasma level was rather high initially, 56 μmol/L (10 mg/L) (page 909), which would correspond to a dietary intake of 100 mg/day or more (Levine 1996). After six weeks, the vitamin C level was 77 μmol/L (+36%) in the vitamin C supplemented group and 52 μmol/L (-7%) in the placebo group (page 909).
The paper by Mochalkin 1970 is in Russian and a translation into English is available. The selection criteria for the participants were not described; neither were many other methodologically relevant aspects. This is a three-arm trial with one control arm and two vitamin C arms with different doses. Placebo was not mentioned and probably was not used in the control arm. However, participants in two other trial arms were administered different doses of vitamin C and the lower-dose arm was used as the reference group in the primary analysis of this review because it seems unlikely that the difference between these arms might be explained by the placebo effect. "The group of patients comprised 140 males diagnosed with acute pneumonia hospitalised during the first two days of onset of the disease [124 patients were 20 to 60 years of age, and 16 were over 60 years]. Depending on the mode of basic treatment, the patients were divided into three groups: Group I (70 patients) was treated with antibiotics without ascorbic acid (25 patients were treated with penicillin, 15 with streptomycin, 15 with penicillin and streptomycin, and 15 with tetracycline); Group II (39 patients) was treated with antibiotics combined with vitamin C (50 mg per 100,000 antibiotic units) (15, 8, 8, 8 patients in the antibiotic groups, respectively); Group III (31 patients) was treated with antibiotics combined with ascorbic acid (100 mg per 100,000 antibiotic units) (10, 7, 7, 7 patients in the antibiotic groups, respectively)" (page 18). Ascorbic acid powder was taken orally. Both antibiotics and ascorbic acid were used for 10 days . . . All patients were tested under equal conditions of placement, care, and nutrition, and were subjected to a complex therapy which included antibiotics . . . To monitor the effectiveness of the employed methods of treatment, we used the following parameters: dynamics of temperature normalisation, erythrocyte sedimentation rate, leucocyte quantity in the peripheral blood, timing of wet rattle disappearance, duration of roentgenologically-determined changes in the lungs, and the mean period of recovery" (page 18). At baseline, the mean plasma vitamin C level was 41 μmol/L. After 10 days treatment, the vitamin C level was 43 μmol/L (+7%) in the higher-dose vitamin C group and 35 μmol/L (-15%) in the low-dose vitamin C group but only 23 μmol/L (-44%) in the control group.
The Kimbarowski 1967 trial was poorly described. Although published in German, an English translation is available. The main focus of the trial was to examine a chemical test, which is not relevant to the current review. However, as a secondary issue, in their report the authors reported the number of bronchopneumonia cases in vitamin C and control groups after hospitalisation. The trial authors excluded the pneumonia cases from their further study (page 2414). For this review the pneumonia cases are relevant since they occurred after vitamin C supplementation was initiated. Although the pneumonia cases occurred after hospitalisation, they occurred within a week and thus did not fall into the category of nosocomial pneumonia. "The studies were conducted with the use of soldiers almost all of whom were of the same age and received the same diet . . . The diagnosis of influenza was based mainly on the clinical pictures and epidemiological data with serological confirmation in a series of cases involving the type A virus." The geographic location where the trial was carried out, the military institution(s), the hospital in which the trial was carried out and the characteristics of the soldiers were not described. The allocation method was not described but the study arms were of closely similar size (112 versus 114 in the control and vitamin C arms, respectively, before excluding the bronchopneumonia cases) so it is probable that allocation occurred sequentially to the two trial arms. The two arms were well-balanced for severity of the influenza. The number of severe cases was 64 versus 65, moderate cases 26 versus 32 and mild cases 12 versus 14 in the two arms respectively (page 2414); the pneumonia cases were not included in these figures. A placebo was not mentioned in the paper and apparently was not used. Blinding of outcome assessment was not described. However, since pneumonia was a secondary issue in the study, the trial authors did not have reason to compare the number of pneumonia cases between the trial arms. It seems improbable, therefore, that the trial authors had substantial bias in their diagnosis of pneumonia. CXR ("Röntgenoscopie") was explicitly mentioned in the paper as a method that was used. It is probable that the diagnosis of bronchopneumonia was based on the CXR but this was not explicitly stated in the paper.
The Glazebrook 1942 trial was the oldest trial identified. The structure of the paper is quite different from more modern trial reports: "In a large training school under our observation there were some 1500 youths aged 15-20 years. For the most part they were drawn from the lower wage-earning classes . . . The food distribution [at the school] was badly managed. . . Often 8 hr. elapsed between the time the food was cooked and its arrival on the dining tables . . .The total intake of vitamin C varied from about 10 to 15 mg per student per day" (pages 4 to 5). "Pure ascorbic acid powder was added to . . . the morning cocoa, and an evening glass of milk. The mixing was done in bulk in the kitchens before issue. The powder dissolved quickly and easily, and did not alter the appearance or taste of the vehicle" (page 7). We consider that the trial corresponds functionally to a placebo-controlled trial because the participants were unable to identify the treatment, although no inactive powder was added to the food of the control group. "The establishment was divided into seven groups or divisions for administrative purposes. The youths of one division worked as a unit, and occupied certain tables in the dining hall. To some extent each division occupied particular dormitories but this separation was not absolute, and there was a fair amount of mixing of divisions in the sleeping quarters. Sleeping and feeding conditions were, of course, the same for all divisions. Careful records had been kept of the incidence of all infections for 1½ years before the observations described here were begun. In the preceding year there had been an epidemic of tonsillitis, which had affected all the divisions uniformly, so that they could not be regarded as separate units within the larger populations" (page 12). "The observations were made by supplying vitamin C in the form of pure ascorbic acid to one or more divisions. This was considered to be the only practical method of carrying out the observations without introducing unnecessary complications. For example, it was not possible to choose boys at random as it would have been impossible to supply them with vitamin C-treated cocoa or milk in the dining room. With the method actually chosen, all that was necessary was to add vitamin C to the supplies of cocoa or milk serving the tables for the appropriate divisions" (page 12). "Moreover, all of the divisions had a population more or less the same as regards duration of stay in the establishment ('institution age'). Infectious diseases were more common amongst those who had more recently joined the institution" (page 12). "When a youth felt ill he was admitted to Sick Quarters unless his complaint was very mild. . . The admission to and discharge from the hospital was not under our control " (pages 13 to 14). [As to pneumonia:] "These cases were subjected to special investigations by us (X-rays, etc.) to establish certain criteria for the diagnosis" (page 16). However, it was not stated whether the diagnosis of pneumonia was carried out by the trial authors of the paper or the physicians at the Sick Quarters. Although the method of diagnosing pneumonia was not described in detail in the paper, with the given descriptions and the severe pathological processes occurring in pneumonia it seems unlikely that vitamin C treatment would have substantially affected the diagnosis of pneumonia.
Risk of bias in included studies
Two of the trials were double-blinded, placebo-controlled RCTs without serious methodological defects and, from the descriptions, there was appropriate allocation concealment in these trials (Hunt 1994; Pitt 1979). Four other trials had methodological shortcomings of varying degrees, as described in the previous section (Glazebrook 1942; Kimbarowski 1967; Mochalkin 1970; Tanaka 2000) and the possible role of these shortcomings in the interpretation of the study results is considered in the Discussion section.
The risk of bias is summarised in Figure 1.
|Figure 1. 'Risk of bias' summary|
Pitt 1979 describes that marine recruits were assigned randomly to the groups from a list of consecutive numbers randomised in pairs. Randomisation was carried out by individual recruits within each platoon. Since the study was double-blind, allocation was concealed. Hunt 1994 states that patients commenced placebo or vitamin C therapy to which they were allocated on a randomised, double-blind basis, without giving further details. Since the study was double-blind, allocation was concealed. Glazebrook 1942 allocated schoolboys as administrative units of the boarding school. Thus, allocation was not concealed for the researchers but it may have been concealed for the schoolboys since the researchers pointed out that vitamin C was added in the kitchen and it did not alter the appearance or taste of the vehicle (cocoa or milk). The number of participants in the vitamin C and placebo arms of the Kimbarowski 1967 was closely equal suggesting alternative allocation but this is not explicitly stated. The report does not give any basis to assume allocation concealment by the researchers, while no conclusions can be drawn for the participants. Mochalkin 1970 has groups of quite different sizes indicating that it was not randomised. No description is given about the forming of the study groups. Tanaka 2000 does not describe the method of allocation; although they use the term "randomised" it is possible that the groups were formed by the treatment month: "randomisation was performed according to the month of admission" (p.327).
The Pitt 1979 and Hunt 1994 studies were double-blind. Glazebrook 1942 stated that vitamin C was added in the kitchen and it did not alter the appearance or taste of the vehicle (cocoa or milk), indicating that the participants were blinded for vitamin C administration (see Included studies). Glazebrook's description further indicates that the diagnosis of pneumonia was made in the Sick Quarter by physicians who were not involved in the study so that they probably were blinded as to the treatment group (see Included studies). The level of blinding cannot be concluded for the Kimbarowski 1967, Mochalkin 1970 and the Tanaka 2000 trials.
Incomplete outcome data
Pitt 1979 stated that 64 marine recruits (7.4% of the initial 862) were removed from their platoons and did not continue in the trial but there was no difference between the study arms. An additional 123 recruits (14.3% of the initial 862) were excluded from the analysis because they did take their pills but there was no difference between the study arms. One recruit was removed from the vitamin C group because of adverse effect. Thus, 22% of participants were not included in analysis but there was no difference between the study arms. Hunt 1994 states that four patients were excluded because of "incomplete information" without further details and 57 remained for the analysis; the distribution of the excluded patients is not described. The Glazebrook 1942 study was carried out in a boarding school and the report does not indicate that school children might have dropped out from the trial. Kimbarowski 1967 did not describe any drop-outs before pneumonia was diagnosed. The Mochalkin 1970 study was carried out in a hospital in the former Soviet Union. No comment on drop-outs is given in the report. However, the distribution of the usage of four different antibiotic treatments is given in the three arms and these groups are identical in size within the arms which suggests that they were planned and that there were no drop-outs. In the Tanaka 2000 trial, there is no indication of drop-outs.
Tanaka 2000, Pitt 1979, Kimbarowski 1967 and Glazebrook 1942 considered pneumonia either of secondary interest or as a nuisance and therefore the findings for pneumonia were not selectively reported because of the type of findings. Hunt 1994 was specifically interested in the treatment of pneumonia but there are no indications in the paper that the reported outcomes would have been selected from a larger set of outcomes. Mochalkin 1970 measured temperature, erythrocyte sedimentation rate, leukocyte level, time of wet rattle disappearance, time of normalisation of CXR and the mean period of recovery. All these outcomes were reported.
Other potential sources of bias
See the section Included studies above.
Effects of interventions
Preventing community-acquired pneumonia
1. Incidence of pneumonia
Three trials reported the number of pneumonia cases in the vitamin C and control groups. All these trials found an 80% or greater decrease in the incidence of pneumonia in the vitamin C group (Figure 2). Since the number of cases in the vitamin C groups was very low (zero to two cases in all of the trials) we used the Peto method for calculating the odds ratio (OR) as an approximation to the risk ratio (RR).
|Figure 2. Prophylactic effect of vitamin C against pneumonia in three trials|
The confidence intervals (CI) in the three trials were wide and overlapped substantially and there is no evidence of statistical heterogeneity (Chi
We used the Peto OR method to calculate an estimate of OR and its 95% CI in Figure 2. However, with only a few observed cases, the exact 95% CI is more appropriate than the approximate 95% CI levels: for Glazebrook 1942, exact OR = 0.0 (95% CI 0.0 to 0.80; mid-P = 0.012), for Kimbarowski 1967, OR = 0.18 (95% CI 0.02 to 0.82; mid-P = 0.018) and for Pitt 1979, OR = 0.15 (95% CI 0.0 to 1.04; mid-P = 0.044). The combined mid-P value (2-t) for the three trials was 0.00004 (Hemilä 1997c), indicating that the differences between the vitamin C and control arms in these three trials were unlikely to be explained by random variation.
Subgroup and sensitivity analyses
We carried out sensitivity analysis in this set of prophylactic trials by excluding trials that did not use randomisation and placebo. This left Pitt 1979 as the only trial with high methodological quality. Nevertheless, the findings of the Pitt 1979 trial did not differ from the other two trials. As noted above, the trials were clinically heterogeneous and we do not expect the same treatment effect in such variable conditions. However, there was no evident trend for the most positive findings to occur in methodologically less satisfactory trials.
Subgroup analysis by vitamin C dosage of less or more than 100 mg/day did not reveal any effect of the dose. However, the trials were clinically so heterogeneous and the number of cases so low that we could not make any conclusions about dose-dependency. All three trials mentioned the usage of the chest radiograph (CXR) but none of them provided a well-defined case definition of pneumonia. Thus we did not carry out a subgroup analysis by use of a CXR for diagnosis.
In the Glazebrook 1942 trial, allocation to treatment groups was carried out by institute 'divisions' and not on the basis of individual boys. Therefore, we also analysed the Glazebrook 1942 trial using the 'division' as the unit of observation. Distribution of pneumonia cases in the five control divisions was 5, 3, 2, 4 and 3 (mean 3.4 cases per division) and in the two vitamin C divisions it was 0 and 0. We assumed that the mean of the control divisions was a suitable estimate for the Poisson distribution mean and used that assumption as a basis for statistical analysis. The size of the individual divisions was not stated in the paper but the two vitamin C divisions had on average 167 boys (335/2) and the five control divisions 220 boys (1100/5), thus the size of the vitamin C divisions was 0.76 times the size of the control divisions. We adjusted the mean incidence by this ratio, so that we expected 2.6 pneumonia cases per vitamin C division, assuming the same incidence as for the control divisions. With this Poisson mean, we calculated the probability that there was no case of pneumonia in two separate vitamin C divisions as having a P value of 0.006. Accordingly, using a 'division' as the unit of observation does not change the conclusions.
Treating community-acquired pneumonia
1. Duration and severity of pneumonic episodes
As a measure of pneumonia severity, Hunt 1994 used 'total respiratory score', which had a range from three (least ill) to nine (most ill) and 10 (dead). On this pneumonia severity score, vitamin C caused a marginally significant benefit at four weeks: P = 0.053, which is based on a decrease in the respiratory score: -2.31 (standard deviation (SD) 2.44) versus -3.43 (SD 1.77) in the placebo and vitamin C groups, respectively. However, the benefit was restricted to patients who were most severely ill when admitted to the hospital (see subgroup analysis below; Figure 3).
|Figure 3. Effect of vitamin C on the respiratory symptom score (scale 3 to 10) in Hunt 1994|
Mochalkin 1970 had three trial arms: control, low vitamin C and high vitamin C. The control arm was not administered a placebo. Therefore, we restricted our main analysis to the comparison of the two vitamin C arms so that the low vitamin C group served as the placebo group in the comparison (Figure 4). Because of Mochalkin's protocol, the mean vitamin C dose of the higher-dose arm was double that of the lower-dose arm, although the dosage ranges within both vitamin C arms overlapped (see Characteristics of included studies). There was a statistically highly significant decrease in the duration of pneumonia in the high vitamin C dose arm compared with the low dose arm (Figure 4; P < 0.0001).
|Figure 4. Effect of vitamin C on the duration of hospital stay (days) in Mochalkin 1970|
2. Death caused by pneumonia
Hunt 1994 found an 85% lower mortality in the vitamin C group compared with the placebo group but this comparison was based on six cases only ( Analysis 2.2; mid-P = 0.12). All deaths occurred in the patients who were most severely ill when admitted to the hospital and all deaths were caused by the respiratory infection.
Subgroup and sensitivity analyses
Hunt 1994 published the total respiratory score values for all participants and for the most severely ill participants (baseline score eight or nine). We calculated the mean scores in the less ill patients (baseline score seven or less). There is significant heterogeneity in the effect of vitamin C on the more ill and the less ill patients with P = 0.02 and I
We had planned a subgroup analysis of therapeutic trials by vitamin C dosage: less and more than 1 G/day. Hunt 1994 used only 0.2 G/day. One of the Mochalkin 1970 arms was lower than 1 G/day but the other arm had a range over 1 G/day and the planned subgroup analysis was thus not possible. However, the secondary analysis of the Mochalkin 1970 study suggests dose-dependency ( Table 1). The duration of recovery was reduced from 23.7 days in the control group by 4.6 days (19%) in the low-dose vitamin C arm and by 8.6 days (36%) in the high-dose vitamin C arm. Since the mean vitamin C dose in the high vitamin C arm was exactly twice the mean of the lower vitamin C arm, the linearity in dose response is striking ( Table 1).
Sensitivity analysis based on the rejection of trials which were not randomised left the Hunt 1994 trial as the only trial with high quality methodology. Thus, here too there was no evident trend that positive findings might be simply explained by methodological shortcomings of the trials. Both therapeutic trials used CXR when evaluating patients but neither provided a well-defined case definition of pneumonia; nor of lower respiratory tract infection in the Hunt 1994 trial. Mochalkin 1970 used normalisation of CXR as one of their outcomes, which implies that changes in CXR were included in their criteria to define pneumonia.
1. Laboratory findings
C-reactive protein (CRP) levels were not reported in the therapeutic studies. Mochalkin 1970 reported the proportion of participants who had normalisation of erythrocyte sedimentation rate within 16 days ( Table 1).
2. CXR changes and body temperature changes during treatment
As a secondary analysis we present the results of the three arms of the Mochalkin 1970 trial in Table 1. The control arm did not receive placebo. Mochalkin reported the proportion of participants who had normalisation of erythrocyte sedimentation rate, who had no fever after seven days and who had normalisation of the CXR in 10 days. For each outcome, the vitamin C arms fared significantly better than the control arm. The number needed to treat to benefit (NNTB) was around two to five for the three outcomes compared to the control group ( Table 1).
Preventing hospital-acquired pneumonia
1. Incidence of pneumonia
Tanaka 2000 reported the incidence of pneumonia within two weeks in severely burned patients after one day of vitamin C administration. No difference between the groups was seen ( Analysis 2.1). However, vitamin C levels had returned to normal within three days and the confidence intervals are wide; the study is therefore not informative.
Treating hospital-acquired pneumonia
We did not identify any trials giving information on the effect of vitamin C in treating hospital-acquired pneumonia.
Summary of main results
We identified three prophylactic trials which recorded 37 cases of community-acquired pneumonia in 2335 people, two therapeutic trials involving 197 community-acquired pneumonia patients and one prophylactic trial recording 13 cases of hospital-acquired pneumonia in 37 patients. The studies on community-acquired pneumonia found a benefit of vitamin C but the trial on hospital-acquired pneumonia did not. One prophylactic and one therapeutic trial on community-acquired pneumonia were satisfactorily randomised, double-blind and placebo-controlled and thus the benefits were not restricted to methodologically less satisfactory trials. However, the five community-acquired pneumonia trials with positive findings were all carried out in conditions that are far from the ordinary way of life for people living in high-income countries.
Overall completeness and applicability of evidence
We consider that the positive findings of five analysed pneumonia trials (Glazebrook 1942; Hunt 1994; Kimbarowski 1967; Mochalkin 1970; Pitt 1979) are reliable and the studies indicate a biological effect of vitamin C. However, in our view great caution is required in the extrapolation of the findings because of limitations caused by various biological factors, for example, the kind of participants used in the trials and vitamin C amounts in the diet.
Participants in the trials and the incidence of pneumonia
Both Pitt 1979 and Kimbarowski 1967 examined soldiers which means substantially dissimilar living conditions compared with ordinary adults. Furthermore, Kimbarowski's soldiers were hospitalised because of influenza, making them a very special group of people. Glazebrook 1942 studied teenage boys in a UK boarding school during World War II. The age range of the Hunt 1994 patients was from 66 to 94 years, obviously restricting any generalisations towards young people. Mochalkin 1970 included a wide age range of participants but their social and nutritional backgrounds were not described in the paper.
An important feature related to the participant selection in the prevention trials was the very high incidence of pneumonia. Glazebrook 1942 and Pitt 1979 recorded 60 and 120 cases of pneumonia per 1000 person-years in their control arms, respectively and Kimbarowski 1967 reported that 10% of their control arm became sick with pneumonia within one week after hospitalisation because of influenza.
In the ordinary middle-aged Western population, the incidence of pneumonia is one to three per 1000 person-years (Baik 2000; Hemilä 2004b). In contrast, in military recruits much higher rates of pneumonia have been reported. The average incidence of pneumonia in marine and naval recruits in the 1970s was 60 per 1000 person-years in a US study (Pazzaglia 1983) which is of the same magnitude as the pneumonia rates in the Glazebrook 1942 and Pitt 1979 trials. Thus, the high incidence of pneumonia makes the conditions of all the three prevention trials very special and limits generalisations of their results. Nevertheless, the consistency in positive findings indicates that vitamin C may influence pneumonia risk in some groups of people.
Vitamin C dose in diet and supplements
An issue of great importance in the interpretation of vitamin C supplementation trials is the level of vitamin C intake in the diet and in the supplements. A different outcome between vitamin C and control arms may result from a very low dietary intake in the control arm ('marginal vitamin C deficiency') or from the high-dose supplementation in the vitamin C arm. In the former case, a small dosage of supplement might produce a similar effect, whereas in the latter case the large dose is essential. As reference levels, scurvy may be caused by vitamin C intakes of less than 10 mg/day, whereas the mean vitamin C intake in the USA is about 100 mg/day (IOM 2000). Previously, a low dietary vitamin C intake level was proposed to explain the reduction in common cold incidence by vitamin C in a set of trials with UK males receiving vitamin C supplementation (Hemilä 1997b; Hemilä 2006).
Glazebrook 1942 estimated that their participants got only 10 to 15 mg/day of vitamin C in their diet, so that the baseline intake was close to scurvy levels. Kimbarowski 1967 and Mochalkin 1970 carried out their studies in the former Soviet Union and it seems highly unlikely that their diet was rich in vitamin C. Hunt 1994 reported overall low plasma levels of vitamin C and the benefit of vitamin C was restricted to the most ill patients who had particularly low vitamin C levels at the baseline: only 20 μmol/L (Figure 3). In the Mochalkin 1970 trial, plasma vitamin C level dropped to 23 μmol/L during the hospital stay in the control group, whereas vitamin C level remained at 43 μmol/L in the high vitamin C dose group. Thus, in these trials control participants may have been suffering from 'marginal vitamin C deficiency'. Low vitamin C levels are not rare in Western hospital patients nor in the Western general community (for example Mosdøl 2008; Raynaud-Simon 2010; Schleicher 2009). Thus, if pneumonia risk is increased by low intakes of vitamin C, this issue may be important also in high-income communities and not just in low-income countries.
The explanation based on 'marginal deficiency' is not applicable to the Pitt 1979 trial with US marine recruits, which reported high baseline vitamin C levels. In the Pitt trial, the baseline plasma vitamin C level was 56 μmol/L, which corresponds to a dietary intake of 100 mg/day or more (Levine 1996). Thus, low dietary vitamin C intake does not explain the benefits of supplementation in the Pitt 1979 trial. Furthermore, this trial used the highest vitamin C dose of the five trials: 2 G/day.
The explanation of 'marginal deficiency' is also not applicable to the comparison between the two vitamin C arms of the Mochalkin 1970 trial. The dose-dependency over the three arms of the Mochalkin trial indicates that the therapeutic effect of vitamin C supplementation was not just limited to treating 'marginal deficiency' ( Table 1). If the vitamin C effect simply alleviated the marginal deficiency, we would not expect difference between the two vitamin C doses. An indication of linear dose-dependency up to 6 G/day of vitamin C was also found in a common cold trial by Karlowski 1975 (see also Hemilä 1999a; Hemilä 2006).
Participants of the Pitt 1979 trial were marine recruits in a training camp, which means heavy physical activity. There is much evidence that physical activity increases the generation of reactive oxygen species (Powers 2008) and, in an electron spin resonance study, the administration of vitamin C prevented exercise-induced oxidative stress (Ashton 1999). Therefore an increased intake of vitamin C might be beneficial for physically stressed people. In support of this view, vitamin C halved the risk of the common cold in five trials with participants under heavy acute physical stress (Hemilä 1996; Hemilä 2013a) and halved the exercise-induced FEV
Thus, it is probable that the heavy physical training of the military recruits of the Pitt 1979 trial is the reason why the high-dose vitamin C supplementation was beneficial for the participants of that trial, but the findings should not be extrapolated to a sedentary population. Several trials with military personnel have found benefit of vitamin C against respiratory infections, which may be explained by heavy physical activity and also by crowded accommodations (Hemilä 2004a).
Diagnosis of pneumonia
Hunt 1994 combined the cases of acute bronchitis and pneumonia together. In young people, acute bronchitis usually has a viral aetiology, whereas the majority of pneumonia cases are caused by bacteria. However, Hunt 1994 patients were all over 60 years of age and their acute bronchitis was "often acute exacerbation of chronic bronchitis", implying bacterial aetiology. The clinical definition of pneumonia is soft and chest X-ray (CXR) has a substantial proportion of false negatives. For such reasons the combined outcome used in the Hunt 1994 trial was appropriate in the current review.
There is no unambiguous clinical definition of pneumonia and this evidently limits the applicability of any studies on pneumonia. See Appendix 1.
Tanaka 2000 administered vitamin C for one day only but recorded pneumonia cases for two weeks. Thus, the follow-up was substantially longer than the vitamin C administration. Furthermore, the confidence interval is wide and does not refute benefits. We consider that this trial is not informative about the potential effects of vitamin C on hospital-acquired pneumonia.
Quality of the evidence
In this section we focus on the five trials with positive findings; we briefly comment on the Tanaka 2000 trial at the end of this section.
Two of the five positive trials were placebo-controlled, randomised trials, whereas the other three trials were technically deficient to varying degrees. Here we consider whether potential biases could explain the differences between the vitamin C and control groups.
The concept of publication bias is based on an assumption that researchers tend to report a study if the result is 'positive' and tend to leave it unreported if the result is 'negative'. With this reasoning, it might be possible that the five trials analysed in this review were published just because of the significant benefit of vitamin C, whereas there might be several trials unpublished because of their negative results. However, the three papers reporting on the prophylactic effect of vitamin C were published separately from the vitamin C effects on pneumonia; the effect on pneumonia was not the motive for publication. Glazebrook 1942 was mainly interested in the common cold and tonsillitis and the effect on pneumonia was mentioned as a secondary issue, indicating that this finding was not the reason for publication. Kimbarowski 1967 considered pneumonia as a nuisance in their trial as they focused on a chemical test. They did not pay any attention to the substantial difference in the occurrence of pneumonia in the trial arms and, for example, in their summary the pneumonia cases in both trial arms were combined. Pitt 1979 focused on the common cold and pneumonia was a secondary outcome which was reported in the text but not in the abstract. Thus publication bias cannot explain this set of three prophylactic trials. Low interest in the effect of vitamin C on pneumonia is also relevant when considering detection bias (see below).
Hunt 1994 found that the effect of vitamin C was limited to the patients with the lowest vitamin C levels and Mochalkin 1970 found a linear dose-response relation in the two vitamin C arms compared with the control group ( Table 1). Neither Hunt nor Mochalkin put proper emphasis on these findings and thus they were not the reason for publication. Publication bias is not a reasonable explanation for the reported positive findings.
Selection bias means that there are systematic differences in the compared groups at baseline. In prophylactic trials, there is a low possibility of bias caused by baseline differences between the treatment arms. Maldistribution of a strong risk factor, such as smoking in a study on lung cancer, would lead to erroneous conclusions on less important risk factors. However, cohort studies have not identified strong risk factors for community-acquired pneumonia, with the age of the person being the most important factor (Baik 2000; Hemilä 2004b). Thus, to explain an 80% or greater reduction in the incidence of pneumonia in the vitamin C arms (Figure 2) there should be spectacular maldistribution in a strong risk factor for pneumonia. Furthermore, the Pitt 1979 trial was randomised and double-blind and Glazebrook 1942 used pre-formed divisions and explicitly considered that the groups of were similar. In the Kimbarowski 1967 trial, the severity of influenza probably was the most important risk factor for the occurrence of pneumonia but it was distributed evenly between the trial arms.
In therapeutic trials, the baseline severity of disease is a factor of obvious importance. The Hunt 1994 trial was randomised and allocation was concealed. Furthermore, the distribution of 'acute bronchitis' and 'bronchopneumonia' and the proportion of 'most severely ill' were closely similar in the treatment arms. Mochalkin 1970 did not describe the distribution of pneumonia severity but antibiotic treatments were distributed evenly in the three arms, so that if the selection of antibiotics depended on the clinical symptoms they were also divided evenly. Thus, it seems unlikely that systematic baseline differences between the trial arms would explain the benefits observed in the vitamin C arms.
Performance bias means systematic differences in the care provided, apart from the intervention being evaluated. The Hunt 1994 and Pitt 1979 trials were double-blinded. According to the Glazebrook 1942 description, the boys in different divisions were treated equally. Kimbarowski 1967 stated that the participants received the same diet which is obviously essential in a vitamin C study but otherwise the similarity of other treatments was not mentioned. Mochalkin 1970 stated that all patients were tested under equal conditions of placement, care and nutrition and, as noted above, the use of antibiotics was similar in the treatment arms. Although Mochalkin did not use a placebo in the control group, performance bias does not reasonably explain the difference between the two vitamin C arms (Figure 4; Table 1). Furthermore, given the significant difference between the two vitamin C groups, it is not reasonable to assume that the difference between the control group and the low-dose vitamin C group is caused by factors other than vitamin C. Such an explanation would presuppose that low vitamin C levels are ineffective and there is a threshold dose over which vitamin C starts to be effective. Such a dose-response model would be opposite to the findings of many studies, indicating that in many cases the benefits are more pronounced in the low-dose region (see above section 'Vitamin C dose in diet and supplements').
Thus, in two trials there was good evidence that participants were treated equally except for the vitamin C supplementation and in the other trials there was no explicit reason to assume that the other treatments would substantially differ between the trial arms.
Attrition bias means high or divergent drop-out proportions and does not seem to be a substantial concern in these five trials. The three trials examining the preventive effect of vitamin C were carried out within military organisations or in a boarding school. Such a background and the descriptions in the papers did not suggest a considerable drop-out problem. Pitt 1979 stated that 22% of the initial population were removed from their platoons or did not continue to take their pills and were not included in the final analysis, but the drop-outs were distributed evenly in the treatment arms. Hunt 1994 followed up the patients for four weeks and did not report any drop-outs. Mochalkin 1970 did not comment on drop-outs but the distribution of antibiotic usage was even in the trial arms, which would seem to exclude any drop-outs.
Detection bias means systematic differences in outcome assessment. The Hunt 1994 and Pitt 1979 trials were double-blinded and bias caused by the knowledge of participants or investigators was unlikely to have affected the outcome assessment. As noted above, in the Glazebrook 1942, Kimbarowski 1967 and Pitt 1979 trials, pneumonia was a secondary issue and it is unlikely that under such conditions the investigators would have any tendency to diagnose pneumonia differently in the trial arms. The Mochalkin 1970 report did not allow any direct or indirect conclusions on the possibility of detection bias.
Even though three of the trials with positive results were methodologically less satisfactory in comparison with modern trial standards, the positive findings of these three trials are not easily explained by biases.
Tanaka 2000 trial was poorly described and the allocation method and level of blinding are not clear from the study report. Nevertheless, we would not ignore the trial on the basis of these methodological shortcomings. Vitamin C was administered for one day only but the pneumonia cases were recorded for two weeks. In addition the confidence interval is wide and does not preclude benefits. We consider that these issues make the study uninformative about the role of vitamin C on hospital-acquired pneumonia.
Potential biases in the review process
Our search of databases for trials meeting the criteria for our review was exhaustive but we also read reference lists of several reviews, such as Briggs 1984, which contained 413 references to papers related to vitamin C and infections. Although there might be unpublished trials or trials published in very difficult to access journals or books, it seems unlikely that we could have missed large controlled trials.
Agreements and disagreements with other studies or reviews
Although the proponents of evidence-based medicine argue that "if you find that the study was not randomised, we'd suggest that you stop reading it and go on to the next article" (Sackett 1997) we consider that cohort studies can give an important perspective to the relation between vitamin C intake and pneumonia risk.
A cohort study found no association between vitamin C intake and community-acquired pneumonia in middle-aged men in the USA (Merchant 2004). However, there are substantial differences between this cohort study and the three prophylactic trials in Figure 2. Merchant 2004 investigated male US health professionals of 40 to 75 years of age, the selection of which meant a population with a much greater than average interest in factors that affect health and whose working conditions are quite sedentary. In Merchant's cohort, the median vitamin C intake of the lowest quintile was 95 mg/day and the overall median was 218 mg/day, whereas the overall median of the ordinary US population is about 100 mg/day (IOM 2000). Thus vitamin C intake in Merchant's cohort was substantially higher than in the three prophylactic trials analysed in this review and the living conditions were very different compared with the three trials. Furthermore, pneumonia incidence was three cases per 1000 person-years. Thus, the Merchant cohort study indicates that the level of vitamin C intake does not affect pneumonia risk in sedentary, health-conscious, middle-aged, upper class populations when the dietary vitamin C intake is over 100 mg/day. Still, their findings cannot be extrapolated to substantially different population groups such as those in the three prophylactic trials analysed in the current review. Nevertheless, the Merchant cohort study sets certain limits to the putative generalisations of the analysed trials. Thus, we consider that biological differences, rather than methodological differences, are the most appropriate explanations for the divergence in the role of vitamin C in Merchant 2004 and the intervention trials in Figure 2.
Some further trials also are relevant to the current topic. Dahlberg 1944 reported respiratory infections which were more severe than the common cold in military recruits (five in the vitamin C group, 10 in the control group of the same size; risk ratio (RR) 0.5, 95% confidence interval (CI) 0.2 to 1.5) but their outcome included otitis and sinusitis and not just lower respiratory infections.
Nathens 2002 reported that in critically ill intensive care unit (ICU) patients vitamin C and E combination had no effect on pneumonia risk (RR 0.79, 95% CI 0.53 to 1.20). However, patients in the vitamin C and E group required 0.9 days (95% CI 0.6 to 1.2 days) less mechanical ventilatory support and had a 1.2 day (95% CI 0.81 to 1.5 days) reduction in their ICU length of stay, indicating benefits of antioxidant supplementation. However, the role of vitamin C per se is ambiguous and the benefit was for an outcome not directly linked to infection, even though a longer ICU stay is associated with a greater risk of pneumonia. In their trial with severe burn patients, Tanaka 2000 found a 43% (P = 0.03) reduction in the duration of mechanical ventilation and a significant decrease in infusion fluid volumes from one-day, large-dose vitamin C administration, even though the incidence of hospital-acquired pneumonia was not influenced ( Analysis 3.1).
Mahalanabis 2006a reported that 400 mg vitamin E and 200 mg vitamin C per day had no therapeutic effect on 2 to 35-month old children with severe acute lower respiratory infection. However, the vitamin E dose was very high for children of this age; for example in the ATBC Study it was 50 mg/day for middle-aged males (Hemilä 2004b). Thus, Mahalanabis 2006a does not allow any specific conclusions on the possible role of vitamin C.
Safety of vitamin C
Pitt 1979 administered 2 G/day of vitamin C to 331 participants for two months. None of the reported symptoms that participants thought to be caused by the pills were statistically more frequent in the vitamin C than in the placebo arm. Urticaria developed in one recruit in the vitamin C arm which subsided when the pills were withheld and recurred when he resumed taking his pills. He was instructed to stop taking pills and was excluded from the final analysis. The other trials used lower vitamin C doses and are uninformative on the safety of high vitamin C doses.
In general, vitamin C is considered safe in doses up to several grams per day and although there has been speculation about potential harms of large doses this has been shown to be unfounded (Hathcock 2005; Hemilä 2006; Rivers 1987). For example, in a pharmacokinetic study, participants were administered up to 100 G of vitamin C intravenously within a few hours without any reported adverse effects, indicating the safety of such a large dose in healthy people (Padayatty 2004). Cathcart 1981 reported that he had administered orally over 100 G per day of vitamin C to pneumonia patients, which indicates the safety of such high doses for pneumonia patients, although such an uncontrolled observation does not provide evidence of benefit. Large doses of vitamin C have also been administered intravenously for numerous patients without adverse effects (Padayatty 2010).
Two large-scale trials with 8171 female health professionals and 14,641 male physicians found no adverse effects of 0.5 G/day of vitamin C when administered for eight to nine years, indicating the long-term safety of such a dosage level (Cook 2007; Sesso 2008).
There are few reports of severe harm caused by high-dose vitamin C administration and the death of a 68-year old African American man was not attributed to intravenous injection of 80 G of vitamin C on two consecutive days per se but to his coincident glucose-6-phosphate dehydrogenase (G6PD) deficiency (Campbell 1975). Such isolated unfortunate events have no public health importance but they indicate that people with G6PD should avoid large doses of vitamin C.
Mechanism of effect
Our review is largely based on the concept that vitamin C affects the immune system, which explains the protection from infections in animals (Hemilä 2006). Such immune system effects are plausible explanations for the benefits observed in the prophylactic trials (Figure 2). However, vitamin C also has non-immune effects that may be relevant in therapeutic trials.
Vitamin C participates in the synthesis of norepinephrine and a series of neuropeptides (Rice 2000) and carnitine which participates in energy metabolism (Hughes 1988; Jones 1982). In a study of experimentally induced vitamin C deficiency, Kinsman 1971 compared high and low levels of vitamin C in whole blood (93 μmol/L and 25 μmol/L) and found that "scores in the neurotic triad of the Minnesota Multiphasic Personality Inventory (the hypochondriasis, depression and hysteria scales) became elevated as deficiency of vitamin C progressed". Therefore, it is possible that in a therapeutic setting the influence of vitamin C is not limited to the immune system. Vitamin C levels in whole blood are higher than plasma levels and thus Kinsman's levels cannot be directly compared with the low plasma levels reported by Hunt 1994 and Mochalkin 1970 (ca 20 μmol/L). Still, low vitamin C levels might cause psychological symptoms, for which vitamin C supplementation might be beneficial. Some of the early case reports of pneumonia patients described particularly rapid benefits of vitamin C (Bohnholtzer 1937; Dalton 1962; Hochwald 1937; Klenner 1948) and such rapid benefits might be caused by non-immunological effects of vitamin C, rather than by immunological mechanisms.
Consistent with the concept that vitamin C might have an influence on general well-being, a recent study reported that vitamin C administration improved the mood of acutely hospitalised patients (Zhang 2011). It is noteworthy that in the Mochalkin 1970 trial the vitamin C level dropped by 44% in 10 days in the control group ( Table 1), consistent with other studies that have found reductions in vitamin C levels with infections (Hemilä 2006). Neither Hunt 1994 nor Mochalkin 1970 measured any index of general well-being or psychological status.
Implications for practice
Vitamin C is relatively cheap and it is safe in doses of grams per day. Nevertheless, with the current evidence there is no basis for the prophylactic use of vitamin C to prevent pneumonia because it would require continuous supplementation with poorly understood effects.
While waiting for new trials, therapeutic vitamin C supplementation may be reasonable for patients with pneumonia who have low vitamin C plasma levels, since therapeutic administration is limited in time. With the low price of vitamin C, the cost-benefit ratio may be reasonable even if the benefit might be substantially lower than that observed in the therapeutic trials analysed in this review.
Implications for research
The incidence of pneumonia is low in the middle-aged population in Western countries (1 to 3 per 1000 person-years) and there is no rationale for studying the prophylactic effect of vitamin C in such a population. Even if vitamin C might have a small effect, the low baseline incidence would lead to a very high number needed to treat to benefit (NNTB).
Certain populations have a high risk of pneumonia. In low-income countries the incidence of pneumonia in children has been up to 400 cases per 1000 person-years (Paynter 2010). In many low-income countries the prevalence of malnutrition is high, indicating low vitamin C intakes (Hemilä 2007a). Elderly people also have an elevated risk of pneumonia, since the incidence increases with age (Baik 2000; Hemilä 2004b). A further population group with a particularly high risk of pneumonia is military recruits (Pazzaglia 1983).
Even if the benefit of vitamin C was substantially lower than in the three prophylactic trials analysed in this review, the effect may still be important in populations with a high incidence of pneumonia. For example, with a baseline pneumonia incidence of 60 per 1000 person-years, a reduction of risk by half would correspond to a NNTB of 33 over one year of such a high risk.
In the USA, pneumonia is the sixth most common cause of death and the most common cause of infection-related mortality, reflecting its importance (Mandell 2010). Therapeutic trials on vitamin C in pneumonia patients should be carried out, in particular in patients with low vitamin C plasma levels but possibly also in participants with ordinary plasma vitamin C levels. The outcomes of therapeutic trials should include soft outcomes measuring well-being because vitamin C may also have non-immune effects, especially in participants with very low plasma vitamin C levels and pneumonia per se leads to a substantial reduction in vitamin C levels.
English translation of the Gander 1936 paper was kindly arranged by Eva Wintergerst from Bayer Consumer Care AG, Basel, Switzerland. English translations of the Bohnholtzer 1937 and Hochwald 1937 papers were kindly arranged by Angelique Boyer from Proctor and Gamble Co, Cincinnati, OH.
We are grateful to Sharon Gudu, Anwar Merchant, Vladimir Bakaev, Andrey Duntau, Bahi Takkouche, Lize van der Merwe and Paul Glasziou for commenting on the draft protocol, which helped us to rewrite the Methods section.
Finally, the authors would like to thank the following people who commented on the draft review: Ann Fonfa, Vladimir Bakaev, Andrey Duntau, Anwar Merchant, Rob Ware and Chris Del Mar.
Data and analyses
- Top of page
- Authors' conclusions
- Data and analyses
- What's new
- Contributions of authors
- Declarations of interest
- Sources of support
- Differences between protocol and review
- Index terms
Appendix 1. Diagnosis of pneumonia
We did not require that the pneumonia diagnosis was based on chest X-radiography (CXR). CXR is objective in the sense that the picture is permanent. However, it is not an effective test for pneumonia. Syrjälä 1998 reported that of 26 adult pneumonia cases that were diagnosed as pneumonia using high-resolution computed tomography (HRCT), only 18 were identified when using CXR; thus 30% of the HRCT-identified pneumonias were false negatives in the CXR. Doherty 1991 reported that of six cases of pneumonia in children identified at autopsy, only three were identified by a radiologist using CXR. Evidently the sensitivity of CXR is often low. Also, the interpretation of CXR to conclude that a patient has pneumonia is quite subjective. Kappa (K) score is a measure for inter-observer variability, with K = 1 indicating perfect agreement and K = 0 indicating agreement explained by pure chance. In both adults and children, interpretation of a set of CXR to decide whether the patient has pneumonia or not yielded low agreement between two observers: K of 0.4 to 0.5 (Albaum 1996; Bloomfield 1999; Hopstaken 2004; Melbye 1992) and inter-observer variability ranged from 20% to 30% (Kiekara 1996; Young 1994). In patients with chronic obstructive pulmonary disease, Hopstaken 2004 found a K of 0.2 for the diagnosis of pneumonia using CXR indicating very poor agreement between two radiologists. Because of its inconsistency and low sensitivity, CXR is not a 'gold standard' for the diagnosis of pneumonia, even though it is a very popular method. HRCT is more sensitive but rarely available.
Pneumonia can be diagnosed clinically without CXR. In low-income countries, the World Health Organization (WHO) has proposed the use of respiratory rate and chest in-drawing to decide whether children presenting to outpatient clinics with cough or difficulty in breathing have 'clinical pneumonia' (Pio 2003). A multi-centre study facilitated by the WHO evaluated the efficacy of predicting pneumonia from clinical signs and symptoms and found that a combination of respiratory rate, rectal temperature, weight-for-age and a set of other clinical findings accurately predicted pneumonia (WHO 1999a). This study was motivated by the fact that "in low-income countries laboratory facilities to perform tests such as the blood count and CXR are often unavailable and clinical decisions must be made without them". A Cochrane Review analysed whether clinical symptoms and signs might be used to identify pneumonia caused by Mycoplasma pneumoniae but they were not good at identifying mycoplasma pneumonia cases (Wang 2012).
Clinical diagnosis of pneumonia and CXR-based diagnosis of pneumonia have modest disagreement but we do not know how they both compare with HRCT. It is possible that some of the clinical pneumonia cases that are negative in CXR might be positive in HRCT. Thus, the divergence between clinical diagnosis and CXR should not be considered categorically as a measure of error in the former.
Furthermore, CXR does not necessarily add useful information to the clinical diagnosis of pneumonia. Swingler found that "interpretation of CXR did not affect clinical outcome in outpatient children with an acute lower respiratory infection. There are no clinically identifiable subgroups of children within the WHO case definition of pneumonia who are likely to benefit from a CXR" (Swingler 1998). Finally, Dirlewanger 2002 in Switzerland found that of 47 children fulfilling the WHO clinical criteria of pneumonia, 46 children had consolidation or diffuse infiltrate in CXR; thus showing that pneumonia can be accurately diagnosed clinically in high-income countries without the use of CXR. For these reasons we have not limited our review to trials using pneumonia diagnosis based on CXR but have also included trials with clinical pneumonia diagnosis.
Appendix 2. Earlier searches
Search in 2007 version: We searched the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2006, Issue 1), OLD MEDLINE (1950 TO 1965), MEDLINE (1966 to February Week 2, 2006), EMBASE (1974 to March 2006), Web of Science (1945 to February 2006) and reference lists of reviews and articles.
1 vitamin C.mp. or exp Ascorbic Acid/
2 pneumonia.mp. or exp Pneumonia/
3 bronchitis.mp. or exp Bronchitis/
4 2 or 3
5 1 and 4
Search in 2009 version: We searched the Cochrane Central Register of Controlled Trials (CENTRAL) (The Cochrane Library 2008, Issue 4), which contains the Acute Respiratory Infections Group's Specialised Register, Ovid MEDLINE (1950 to January Week 1, 2009), EMBASE (1974 to January 2009), Web of Science (1945 to January 2009) and reference lists of reviews and articles. MEDLINE search as in 2007 version, see above.
Appendix 3. Embase.com search strategy in 2013
#13. #6 AND #12
#12. #7 OR #8 OR #9 OR #10 OR #11
#11. 'vit c':ab,ti
#10. 'vitamin c':ab,ti
#7. 'ascorbic acid'/exp
#6. #1 OR #2 OR #3 OR #4 OR #5
Appendix 4. Web of Science search strategy in 2013
Topic=(pneumon* or bronchopneumon* or bronchit*) AND Topic=("ascorbic acid" or "vitamin c" or "vit c")
Last assessed as up-to-date: 8 April 2013.
Protocol first published: Issue 4, 2005
Review first published: Issue 1, 2007
Contributions of authors
HH planned the review, carried out the literature searches and RevMan analyses and wrote the draft of the text. HH and PL selected the included trials, extracted the data, interpreted the results and finished the text. HH updated the review and PL commented on the update.
Declarations of interest
Sources of support
- No sources of support supplied
- New Source of support, Not specified.
Differences between protocol and review
In our protocol we defined primary outcomes for treatment as: "the duration and severity of pneumonic episode, duration of hospital stay and death caused by pneumonia". However, "the duration of hospital stay" is one way of measuring "the duration of pneumonia" and therefore in the 2013 update we simplified the primary treatment outcomes to: "the duration and severity of pneumonic episode and death caused by pneumonia".
Full text versions of references which are available either free or from the publishers' databases can be accessed via the home page of the contact author, Harri Hemilä: http://www.mv.helsinki.fi/home/hemila/CP .
Medical Subject Headings (MeSH)
Antioxidants [*administration & dosage]; Ascorbic Acid [*administration & dosage]; Community-Acquired Infections [prevention & control]; Pneumonia [*prevention & control]; Vitamins [*administration & dosage]
MeSH check words