The impact of organ failures and their relationship with outcome in intensive care: analysis of a prospective multicentre database of adult admissions
Tonga K. Nfor, MD MSC
The database of a multicentre cohort study was analysed to determine the impact of intensive care unit (ICU) organ failures and their association with ICU mortality using sequential organ failure assessment (SOFA). A consecutive sample of 873 adult patients with a non-neurological diagnosis was identified. SOFA scores were measured every 24 h of ICU stay. The odds of ICU death within 7 days doubled (95% CI 1.3–2.9) for a 5-unit increase in total SOFA score at admission, p < 0.001. However ICU death after 7 days was not associated with total SOFA score at admission, p = 0.36. Compared to patients with a day 6 total SOFA score = 5, there was a 1-unit (95% CI 0.8–3.1) increase in the odds ratio of ICU death after 7 days with every 5-unit increase in SOFA score on day 6, p = 0.009. Continuous assessments of organ failures during an ICU admission are more useful than scores measured at admission to determine outcome and to compare ICUs.
Recent guidelines for admission, clinical management and organ support in intensive care have highlighted the need for accurate characterisation of patterns of illness and outcomes for intensive care populations [1–3]. Organ support in the intensive care unit (ICU) can be usefully classified in relation to respiratory, cardiovascular, renal, hepatic, coagulatory and central nervous dysfunction. Among scoring systems developed for measuring the severity of illness in the ICU, the Sequential Organ Failure Assessment (SOFA) [4, 5] is designed to quantify specifically dysfunction in individual organ-systems using repeat measurements on successive days of the patient's ICU stay . Other scores such as the Acute Physiology and Chronic Health Evaluation (APACHE) [7, 8], and Simplified Acute Physiology Score (SAPS) [9, 10], which are measured on the day of ICU admission, are less suited to the follow up and the evolution of organ failure and are not associated with outcome in long-stay patients .
Although it has been shown that the various organ-systems affect ICU mortality differently , the relative contribution of each system failure is unclear. In addition, few studies have specifically examined the burden of organ failure in well-defined cohorts of ICU patients, or the evolution of organ failure during ICU care. These data could be useful for national service planning or for comparing individual ICUs. A few studies have assessed the relationship between the SOFA score and mortality, but they used the total SOFA as a continuous variable and had limited ability to assess the relative contribution of individual organ systems [12, 13]. Our primary aim in this study was to examine the epidemiology of organ failure in Scottish ICUs based on the alternative approach of modelling SOFA as a categorical variable. We used SOFA to quantify the impact of organ failures at ICU admission and during an ICU admission. Secondary aims were to investigate the relationship between organ failures and ICU mortality using both individual and combined organ system scores, and to explore associations for patients receiving prolonged ICU care.
Local research ethics committee approval was obtained for this study.
This study was based on the database obtained from a prospective multicentre study, the Audit of Transfusion in Intensive Care in Scotland – ATICS, conducted in 2001 [14, 15]. The study was conducted in 10 of the 26 general ICUs in Scotland and included 44% of all admissions in the country during the study period . Cardiac and paediatric ICUs were not included. All patients admitted between 4th June and 12th September 2001 (100 days) were consecutively sampled. The study included > 99.5% of all admissions to the participating ICUs over the study period, and the cohort was shown to have similar demographics and illness severity to admissions to non-participating Scottish ICUs over the same period . For the purpose of this analysis we did not include patients who were < 18 years old at admission or who were receiving palliative care at study admission. We also excluded patients with a primary neurological diagnosis because we considered outcomes for this group were very dependent on the nature and severity of the neurological diagnosis rather than other organ dysfunctions. Only one admission period was analysed for each patient; re-admissions within the study period were excluded.
The predisposing variable for this study was organ dysfunction measured daily using the SOFA score [4, 5]. The organ systems considered were respiratory, cardiovascular, renal, coagulatory and hepatic. Central nervous system dysfunction (measured by the Glasgow coma score) was not considered because most ICU patients received sedation, which significantly limits the validity of the score . Each of the five systems was scored on a discrete scale of 0–4 in order of increasing severity of dysfunction, producing a total SOFA score of 0–20. The SOFA score is not a continuous scale. For the purpose of analysis, severity of organ failure for each system was categorised as absent for score = 0, mild-moderate for scores 1–2 and severe for scores 3–4. Pre-specified levels for total SOFA score were no organ dysfunction for score 0, mild illness 1–5, moderate 6–10 and severe illness > 10.
The outcome was all-cause mortality during ICU stay. Patients were described as short-stay patients if they spent ≤7 days and long-stay if they spent > 7 days in the ICU.
Data were prospectively collected following a standardised protocol by research staff in each unit . A patient's daily SOFA scores were calculated over each 24-h period spent in the ICU from routine clinical assessments [see Appendix 1]. Other clinical data were obtained from the patient's medical records. Demographic data and ICU mortality status were obtained from the Scottish Intensive Care Society Audit Group (SICSAG) database.
The study was based on a fixed dataset of patients so a formal sample size calculation was not mandatory. Patient characteristics were summarised as mean (and standard deviation) if numerical or percentages if categorical.
The impact of organ failures on intensive care admission and during ICU stay
Prevalence of organ dysfunction was calculated as the proportion of all patients who had dysfunction in the organ system of interest (SOFA ≥ 1). We calculated the point prevalence of each organ failure at ICU admission and the period prevalence of organ dysfunction over the total ICU stay for each organ system. Only one episode of dysfunction was counted for each patient. We calculated the incidence of organ dysfunction in each organ system as the proportion of patients who were unaffected at ICU admission (patients at risk; organ score at admission = 0) who developed new organ failure per day spent in ICU. The total burden of organ failures for the ICUs was measured by the total ICU time in patient days occupied with a particular organ-system failure. This was calculated by summing up the respective lengths of time spent by each patient with that organ failure. This can be expressed using the formula for total patient days of organ failure.
The proportion of total ICU time occupied by a particular organ dysfunction was calculated by dividing the total patient days spent with that organ failure by the total patient days of ICU stay in the whole sample.
Relationship between total SOFA score and ICU mortality
Binary logistic regression was used to investigate the relationship between total SOFA score and ICU mortality. The odds of ICU death in patients with a total SOFA score of 6–10 (moderate) and > 10 (severe) at admission was compared to patients with scores of = 5 (mild) at admission. This analysis was done separately to investigate the association with overall ICU death, death in short-stay patients (= 7 days) and death in long-stay patients (> 7 days). Secondary analysis was adjusted for age, sex, source of referral and APACHE III diagnostic category. For patients who spent > 7 days in the ICU, the effect of total SOFA score at day 6 on ICU mortality was also examined as above.
Contribution of individual organ failures to ICU mortality
Multivariate binary logistic regression was used to assess the effect of individual organ systems on ICU mortality. In each of the five systems, patients having a maximum score during ICU stay of 1–2 (mild-moderate dysfunction) or 3–4 (severe dysfunction) were compared with those having no dysfunction (score = 0). Analysis was adjusted for age, sex, source of referral and diagnostic category.
Effect of combined organ dysfunctions on ICU mortality
The importance of combined organ dysfunctions was examined by comparing patients who had only respiratory dysfunction (which was highly prevalent at ICU admission) with those who had respiratory + 1, + 2, + 3 and + 4 other system dysfunctions. Analysis was adjusted for age, sex, source of referral, diagnostic category and total SOFA score at admission.
SPSS 12.0 was used for all analyses.
Characteristics of patients
During the audit period 1042 patients were admitted to participating units (44% of all Scottish ICU admissions in same period). Of those, 169 (16.2%) met the exclusion criteria, mainly because of a neurological diagnosis (133 patients), leaving a final sample of 873 patients. Of the 5913 patient days of follow-up, data were unavailable on 0.3% for the cardiovascular, renal 0.8%, coagulatory 1.0%, respiratory 1.3% and hepatic 3.2% scores.
Table 1 shows patient characteristics. Patient ages ranged from 18 to 91 years with a mean (SD) of 59.6 years (16.7). The most common diagnoses at admission were respiratory, digestive and non-coronary cardiovascular diseases. In all, 12.2% of patients were transferred from a different hospital, 37.5% were admitted from the surgical theatre and the remainder were from other wards in the same hospital.
Table 1. Characteristics of the patients included in the study.
|Age, mean (SD)||59.6 years (16.7)|
|ICU stay; mean* (SD)||2.3 days (9.1)|
|ICU mortality; %||26.6|
|APACHE II score, mean (SD)||17.8 (9.8)|
|SOFA score on day of admission, mean (SD)|
| ||Respiratory 2.8 (0.9)|
| ||Cardiovascular 1.6 (1.6)|
| ||Renal 0.9 (1.2)|
| ||Coagulatory 0.7 (1.0)|
| ||Hepatic 0.5 (0.9)|
| Total SOFA score||6.4 (3.5)|
|Diagnostic categories; % of patients|
| Peripheral vascular||4.8|
| GI bleeding||4.6|
| Other GI||10.5|
| Self-inflicted overdose||4.0|
| Other respiratory||10.3|
| Liver failure/transplant||2.7|
| Septic shock||7.1|
| GI perforation||6.1|
| Trauma (non-CNS)||4.9|
| Other cardiovascular||1.0|
Impact of organ failure
The point prevalence and period prevalence for each organ system dysfunction are shown in Table 2. The most common dysfunctions were respiratory and cardiovascular. Overall, 95% of patients had respiratory dysfunction at the time of ICU admission, mostly severe (75% of admissions). The prevalence of cardiovascular dysfunction was 65%, with 33% of patients having severe dysfunction at the time of admission. Severe dysfunction was less common in the other organ-systems throughout ICU stay.
Table 2. Prevalence of different organ failures at admission and over intensive care unit (ICU) stay, and percentage of total ICU time (5913 patient days) spent with organ failures.
Table 3 shows the rates of development of new organ failures. There were fewer patients at risk (no dysfunction at admission) of new respiratory and cardiovascular dysfunctions because of their high prevalence at admission. Despite this, respiratory and cardiovascular dysfunction had the highest incidence and developed most rapidly. The incidence of new hepatic, renal and coagulation dysfunction was lower and occurred later during the patients' ICU stay.
Table 3. Development of new organ failures among patients unaffected at admission to intensive care.
Relationship between total SOFA score and ICU mortality
Overall ICU mortality
Table 4 shows the progressive increase in overall ICU mortality rate across levels of increasing total SOFA score at admission (Chi-squared test for trend = 39.1, p < 0.001). The odds of death in patients with total SOFA score of 6–10 was more than double the odds in patients with a total SOFA score of 1–5 (Odds ratio 2.3; 95%CI 1.6–3.2). The odds of death in patients with a total SOFA score of > 10 was almost four times higher than in patients with a total SOFA score of 1–5 (Odds ratio 3.7; 2.4–5.8). This association was confirmed after adjustment for age, sex, source of referral and APACHE III diagnostic category.
Table 4. Relationship between total sequential organ failure assessment (SOFA) score at admission and overall ICU mortality, mortality in short-stay (= 7 days) and long-stay (> 7 days) patients. Relationship between total SOFA score on day 6 and ICU mortality in long-stay patients.
|Overall||SOFA score at admission|
|6–10||338||32.0||2.3 (1.6–3.2)||2.0 (1.3–2.9)|
|>10||122||43.4||3.7 (2.4–5.8)||2.9 (1.8–4.7)|
|Short-stay||SOFA score at admission|
|6–10||338||21.0*||1.9 (1.3–2.9)||1.6 (1.03–2.5)|
|>10||122||36.1*||4.1 (2.5–6.6)||3.0 (1.8–5.1)|
|Long-stay||SOFA score at admission|
|6–10||110||33.6||1.6 (0.8–3.1)||1.6 (0.8–3.4)|
|>10||31||29.0||1.3 (0.5–3.3)||1.4 (0.5–4.0)|
|SOFA score at day 6|
|6–10||99||33.3||1.9 (1.0–3.8)||1.7 (0.7–3.8)|
|>10||39||43.6||2.9 (1.3–6.8)||3.4 (1.2–9.9)|
Mortality before and after 7 days
The relationship between total SOFA at admission and outcome was different for patients who died within 7 days compared with those who suffered ‘late’ deaths (≥ 7 days) (Table 4). There was a progressive increase in ICU mortality with increasing total SOFA score on day of admission for patients who died within 7 days in ICU (Chi-squared test for trend = 34, p < 0.001). The odds of ICU death within 7 days doubled for each 5-unit increase in total SOFA score on the day of admission. This relationship did not exist for patients who remained in the ICU for ≥ 7 days (n = 220). For this subgroup, ICU mortality after the 7th day was not associated with SOFA score at admission (Chi-squared test for trend 2.1, p = 0.36) However, when we explored the association between organ failures on day 6 and ICU mortality after 7 days of stay there was a strong association between SOFA score and eventual ICU mortality (Chi-squared test for trend 6.9, p = 0.009; Table 4). This translated into an approximate doubling of the odds of ICU death for a 5-unit increase in SOFA score on day 6 (Table 4). This association remained strong even after controlling for SOFA score on day of admission.
Contribution of individual organ failures to ICU mortality
The association between individual organ dysfunctions and ICU death is shown in Table 5. The SOFA scores used were the maximum recorded for each organ system during ICU stay. The odds of death in patients with ‘severe’ (SOFA 3–4) and ‘mild-moderate’ (SOFA 1–2) dysfunctions were compared to odds of death in patients with no dysfunction (SOFA = 0). The effect of each organ-system was controlled for the effect of all other four systems. The odds of ICU death increased with increasing level of dysfunction for all organ systems, but this was only statistically significant for cardiovascular, respiratory and renal failures. There were no deaths in patients who did not develop respiratory failure in ICU. Although the absolute ICU mortality rate in patients who had severe cardiovascular and renal failures were similar (49.6% and 51.7%, respectively), cardiovascular failure was far more strongly associated with ICU death when odds of death in patients with severe failure was compared with odds in patients with no failure (Odds ratio 33.3 for severe cardiovascular and 4.0 for severe renal failure). Similar results were obtained after adjustment for baseline patient characteristics.
Table 5. Effect of different organ-system failures occurring during ICU stay on ICU mortality.
| ||0†||167||1.6|| 1|| 1|
|1–2||264||10.3||6.9 (1.6–29.5)||5.5 (1.3–24.0)|
|3–4||377||49.6||33.3 (8.0–139.8)||28.6 (6.7–122)|
| ||0||21||0|| 0|| 0|
|1–2†||109||6.9|| 1|| 1|
|3–4||678||31.1||3.6 (1.4–9.7)||2.7 (1.0–7.6)|
| ||0†||351||8.9|| 1|| 1|
|1–2||258||31.6||2.2 (1.4–3.7)||1.7 (1.0–2.8)|
|3–4||199||51.7||4.0 (2.4–6.9)||3.0 (1.7–5.4)|
| ||0†||373||18.8|| 1|| 1|
|1–2||306||27.2||0.9 (0.6–1.5)||1.0 (0.6–1.6)|
|3–4||129||46.6||1.2 (0.7–2.0)||1.4 (0.7–2.5)|
| ||0†||451||18.7|| 1|| 1|
|1–2||286||29.9||0.8 (0.5–1.2)||0.9 (0.5–1.4)|
|3–4||71||56.3||1.5 (0.8–2.9)||1.8 (0.8–3.6)|
Effect of combined organ failures on ICU mortality
Using respiratory failure as baseline, the effect of additional organ failures on ICU mortality was assessed (organ failure defined as SOFA ≥ 1) (Table 6). In all, 98% of patients had respiratory failure at some point during ICU stay. There was a significant increase in ICU mortality with increasing number of organ failures (Chi-squared test for trend 30.4, p < 0.001). There was an average increase in the odds ratio of ICU death of 1.1 for each additional organ failure compared to respiratory failure only. This relationship was confirmed after adjusting for baseline clinical and demographic characteristics. Further analysis of this relationship revealed that patients having multiple organ dysfunctions but no cardiovascular failure had a similar probability of death in ICU to those having only respiratory failure throughout ICU stay (Odds ratio 0.8, 95% CI 0.3–2.0). Conversely, the odds of ICU death in patients who had multiple organ dysfunctions including cardiovascular failure was four times higher than patients with only respiratory failure (OR 4.2, 95% CI 2.2–8.0). These analyses confirmed the strong independent association between ICU death and cardiovascular failure.
Table 6. Effect of multiple organ-system failures on intensive care unit mortality.
|Respiratory + 1 other system||182||21.4||2.3 (1.1–4.7)||2.3 (1.1–5.0)|
|Respiratory + 2 other systems||178||28.7||3.4 (1.7–6.9)||2.3 (1.03–5.3)|
|Respiratory + 3 other systems||216||30.6||3.7 (1.9–7.4)||2.6 (1.2–5.9)|
|All 5 organ systems||139||41.0||5.9 (2.9–12.0)||3.5 (1.4–8.8)|
We have shown that SOFA scores are a useful method of describing the epidemiology of critical illness both at the time of ICU admission and during ICU care. In well-defined cohorts of patients such as the national database examined in this study, this information can provide useful information on the burden of organ failures in geographical regions. It could also be used as an indicator to evaluate the impact of service changes or the introduction of new therapeutic strategies or drugs. We confirmed the association between organ failures and ICU death and provided new analyses suggesting that cardiovascular failure has the most important independent influence on survival.
There are limitations to the generalisability of our findings. The sample size has limited power to detect associations within subgroups. This explains the wide confidence intervals obtained for some estimates. We used only five of the six components of the SOFA score because of the difficulty in obtaining valid daily neurological scores in sedated patients. To limit the bias this could cause we did not include patients with a neurological diagnosis from our analysis. In such cases it is safer to leave out the neurological component of the SOFA altogether than assume a normal score for non-neurological patients, particularly if using data to predict mortality .
By reference to the national Scottish audit database we knew the final sample used for our analysis comprised 37% of the 2367 patients admitted to ICUs in the whole of Scotland during the recruitment period. The characteristics of the patients in our sample were similar to values reported in the annual audit of Scottish ICUs for the same year . We therefore believe that the study sample was representative of the Scottish general non-neurological ICU population. ICU mortality in our sample (26.6%) was slightly higher than the Scottish average (23%). This can be explained by the fact that most of the patients not included in our analysis were either neurological or young patients who have a better prognosis . ICU mortality among excluded patients was 17%.
There was a positive relationship between total SOFA score at admission and ICU mortality before 7 days but there was no relationship with mortality after 7 days. Instead, ICU mortality after 7 days was associated with high SOFA scores on day 6 even after controlling for SOFA score at admission. This suggests that in patients who survive the first week of intensive care the severity of illness at admission becomes less relevant and that it is the evolution or presence of organ failures late in ICU that becomes important. Consistent with this observation, Moreno et al.  showed that ICU mortality was more strongly associated with ‘delta’ SOFA score (total maximum SOFA minus SOFA score at admission) than SOFA score at admission. Hughes et al.  found no significant difference between the APACHE II scores of ICU survivors and non-survivors spending ≥ 30 days in the ICU. This is an advantage of the SOFA, which is simpler to use for daily monitoring of organ failure, over other ICU scores like APACHE , simplified acute physiology score (SAPS)  and mortality probability model (MPM) [20, 21]. Although increasing dysfunction in each organ-system seemed to be associated with ICU death univariately, when the effect of each organ failure was assessed controlling for the effect of the other four organ-systems, only cardiovascular, respiratory and renal failures remained significant. We confirmed the importance of multiple-organ dysfunction on mortality [22, 23] by demonstrating a steady increase in the odds ratio of ICU death with increasing number of organ failures. Our multivariate analysis indicated that the development or presence of cardiovascular failure had by far the strongest independent effect on the probability of death compared to the presence or development of other organ system failures. This finding highlights the fact that SOFA is a descriptive system and that clinicians should not apply equal weight to each organ system component. Our finding confirms the clinical impression that persisting or new cardiovascular failure in the ICU carries a poor prognosis.
To the best of our knowledge no published study has looked at the burden of specific organ failures in intensive care before. The SOFA provides an opportunity to quantify dysfunction in individual organ systems over time in a way that is not possible with most other ICU scoring systems. Knowing the burden that different organ systems impose on intensive care services is vital to service planning, provision of resources and monitoring. Prevalence and number of person days of organ failure could be useful for the allocation of resources and to track trends over time. The use of incidence rates for new organ failures in patients unaffected at admission could be useful for monitoring purposes over time. Official guidelines for intensive care services [2, 18, 19] stipulate that the objective of intensive care is not only to treat organ failure but also to prevent organ failure in predisposed patients. Incidence of organ failure is an indicator of the degree of attainment of this objective. On all measures the respiratory and cardiovascular systems clearly dominated the burden of organ failure in intensive care units. The fact that three quarters of respiratory failure cases were severe shows its importance as a determinant of ICU admission.
In conclusion, respiratory and cardiovascular failures account for the vast majority of the burden of organ failures in Scottish ICUs. Although organ system failures at ICU admission predict ICU death, this relationship is weak for patients subsequently requiring prolonged ICU care. In this group, repeated assessments later in ICU stay is predictive of subsequent ICU mortality. The risk of death increases with increasing number of organ failures. The presence of cardiovascular failure has the strongest individual relationship with ICU mortality compared with other non-neurological organ-system failures. Used in conjunction with outcome data, organ failures at ICU admission and during ICU care can be used to assess burden of organ failure, compare ICUs or evaluate the impact of changes in health care over time.
Conflict of interest
The authors declare no conflict of interest.
Table Appendix1. Calculation of the SOFA score (adapted from Vincent et al.).
| Pao2/FIo2 (kPa)|
|>53.32||≤53.32||≤40||≤27 with |
|MAP||Dopamine* ≤5 |
or dobutamine or
|Dopamine > 5 or adren ≤0.1* or |
noradren > 0.1
|Dopamine > 15 or adren > 0.1 or|
noradren ≤ 0.1
| Creatinine output; μmol.l−1||<110||110–170||171–299||300–440||>440|
| Urine output; ml.day−1||>500||>500||>500||<500||<200|
| Platelets; × 103.mm−3||>150||≤150||≤100||≤50||≤20|
| Bilirubin; μmol.l−1||<20||20–32||33–101||102–204||>204|
| Glasgow coma score||15||13–14||10–12||6–9||<6|