Dr Ewa Pańkowska The Second Department of Pediatrics Medical University of Warsaw ul. Działdowska 1/3 01-184 Warsaw Poland. Tel: +48 22 4523302; fax: + 48 22 4523309; e-mail: firstname.lastname@example.org
Objective: To investigate potential effects of continuous subcutaneous insulin infusion (CSII) compared with multiple daily injections (MDI) on glycemic control in children with type 1 diabetes mellitus (T1DM).
Study design: Meta-analysis and systematic review of randomized control studies (RCTs). The electronic databases MEDLINE, Cochrane Library, and EMBASE were searched through October 2007.
Results: Six RCTs involving 165 participants with T1DM met our predefined inclusion criteria. Combined data from all trials showed that the CSII group compared with the MDI group experienced a significant reduction in the level of glycosylated hemoglobin. The pooled weighted mean difference (WMD) was −0.24% [95% confidence interval (95% CI) −0.41 to −0.07, p < 0.001] with a fixed model and remained significant in the random effect model. This effect was reached by slightly decreasing insulin requirement [three RCTs, n = 74, WMD −0.22 IU/kg/d (95% CI −0.31 to −0.14, p < 0.001)]. No differences in the incidences of ketoacidosis and severe hypoglycemic events were found.
Conclusions: In short-term insulin therapy, CSII compared with MDI is a more effective form of metabolic control and allows reducing the daily insulin requirement. Yet, no conclusions have been made so far whether this effect holds in later years. These results should be approached with caution because of the methodological limitations of the analyzed studies.
The Diabetes Control and Complications Trial (1) provided conclusive evidence that strict glycemic control reduces the development and progression of late complications in patients with type 1 diabetes mellitus (T1DM). Optimal metabolic control requires replacement therapy that mimics the insulin profile seen in non-diabetic people as closely as possible. This form of insulin treatment, also referred to as basal to bolus therapy, can be administered with multiple daily injections (MDI) or with continuous subcutaneous insulin infusion (CSII). The latter treatment is a relevant therapeutic option in toddlers and preschool-aged children (2–5) as well as in adolescents and adults with T1DM (6, 7). The high precision of modern insulin pumps, flexibility, and easier adaptation of insulin dosing for current physiological needs enable to reduce the risk of severe hypoglycemia in children with T1DM (8–10) and improve glycemic control in adults (11, 12).
Three meta-analyses (13–15) published over the past 5 yr have shown the impact of CSII therapy on metabolic control compared with MDI. All confirmed that CSII significantly improves the reduction in the glycosylated hemoglobin (HbA1c) value. They include trials of different study designs (randomized and non-randomized trials, cohort studies, and observational studies), in addition, different age-groups (13, 14) and patients with T1DM, type 2 diabetes, and gestational diabetes (14). Moreover, all of them included studies published more than 5 yr ago and were not dedicated to the pediatric age-group (13–15). No analysis is currently available with the results of randomized control trials (RCTs) conducted in children with T1DM comparing the effect of CSII with MDI on metabolic control as well as the safety and side effects of this treatment. The aim of our study was, therefore, to investigate the potential effects of CSII compared with MDI on glycemic control in children with T1DM.
Inclusion criteria in this review
We included parallel RCTs or randomized crossover studies comparing CSII therapy with MDI with a minimum 8-wk study duration. Participants were children, adolescents, and young adults aged 1–21 yr with T1DM and minimum duration of diabetes of 3 months. The primary outcome measure was glycemic control, which was designated by the value of HbA1c after the third month and the end of the study. The secondary outcome measures were as follows: a change in HbA1c from baseline until the end of the study, total insulin dose per kilogram per day at the end of the study, severe hypoglycemia rates, diabetic ketoacidosis rates, therapy discontinuation rates, z-score body mass index (BMI) at the end of the study, and quality of life (QoL). In addition to these outcomes, we made an a priori decision to extract other data reported by investigators if they were found to be relevant for this review.
The following electronic databases were systematically searched for relevant studies: MEDLINE (1966–January 2007), the Cochrane Database of Systematic Reviews (issue 4, 2006), and the Cochrane Controlled Trials Register (issue 4, 2006), EMBASE (1980–January 2007). This search was updated in October 2007. Furthermore, reference lists from original studies and review articles were identified. No limit was imposed regarding the language of publication, but certain publication types (i.e., letters to the editor, abstracts, and communications from scientific meetings minutes of scientific meetings) were excluded. The text-word terms and MeSH headings used were diabetes, diabetes mellitus type 1, insulin-dependent diabetes, insulin, continuous subcutaneous infusion, continuous subcutaneous injection, CSII, pump therapy, intensive insulin therapy, multiple daily injections, MDI, flexible multiple daily insulin, FMDI, MSI, children, child*, newborn, infant*, teenagers, teenag*, adolescent*.
Titles and abstracts identified in accordance with the search strategy described above were screened independently by two reviewers. All potentially relevant articles were retained, and full texts of the studies were examined to determine which of them satisfied the inclusion criteria. Data extraction was carried out independently by all reviewers using standard data extractions. Only studies published in the form of complete peer-reviewed publications were included. In addition to the electronic search, we checked out cross-references from the original articles and reviews. In the case of studies in which results were expressed in figures, but not in numbers, we contacted the authors for clarification. We contacted multiple authors (16–20), and finally, two researches answered our queries (18, 20). Data from crossover studies for both periods were combined. As both arms were separated by a 2-wk run-in (washout) period, carryover as an effect of treatment during the first period in these two studies was minimized. Discrepancies between the reviewers were resolved by discussion.
The reviewers independently, but without disregarding the authorship or the journal of publication, assessed the quality of studies meeting the inclusion criteria. The use of the following strategies associated with good quality studies was assessed: (i) allocation concealment, (ii) blinding of investigators, (iii) participants, (iv) outcome assessors and data analysts (yes/no), and (v) comprehensive follow-up. Allocation concealment was considered adequate when the randomization method used did not allow the investigator or participants to identify or influence the intervention group before eligible participants were enrolled in the study. However, the quality of allocation concealment was considered unclear when randomization was used, but no information about the method was available and inadequate when inappropriate methods of randomization (e.g., alternate medical record numbers, unsealed envelopes, and tossing the coin) were used. With regard to the intention-to-treat (ITT) analysis, a ‘yes’ answer meant that the authors had specifically reported undertaking this type of analysis and/or that our own study confirmed it was the case. Conversely, a ‘no’ answer meant that authors did not report the use of an ITT analysis and/or that we could not confirm its use at study assessment. To evaluate the completeness of patient follow-up, we determined the percentage of participants excluded or lost to follow-up.
The data were analyzed using review manager 4.2.10 (version date 13 November 2006, The Cochrane Collaboration). The weighted mean difference (WMD) between the treatment and the control groups was selected to represent the difference in continuous outcomes. Because of a limited amount of data, we did not test for publication bias. The difference between study groups was considered significant when the p value was <0.05 or when the 95% confidence interval (95% CI) for WMD did not exceed 0 (equivalent to p < 0.05). The weight given to each study was based on variance inversion. We assessed heterogeneity between trials using the chi-squared test and I2, which represents the percentage of total variations between studies that are attributed to heterogeneity rather than chance.
We initially selected eight articles (16–23). Table 1 summarizes the characteristics of trials qualified for review. Six RCTs (17–19, 21–23) involving 165 participants (81 in the experimental group and 84 in the control group) met our predefined inclusion criteria. The two remaining studies were excluded (16, 20). In brief, one study was excluded because it was not RCT. The second excluded trial was RCT but performed in children and adults and input data for children were not analyzed separately.
Table 1. Characteristics of included trials
Generation of randomization scheme
Blinding (yes/no/not reported)
Age (yr ± SD) CSII/MDI
Duration of intervention (months)
CSII, number of patients included (finished study)
Adequate (randomization scheme, random number table with block size of four)
12.5 ± 3.2/13 ± 2.8
The mean age of participants varied from 1.7 to 21 yr. The duration of the intervention ranged from 6 to 12 months. Two studies were crossover trials (19, 20).
The trials differed in methodological quality (Table 1). Three trials used an adequate method of generating the randomization scheme; the method used in the remaining four trials was unclear. All trials were described as open-label trials. An adequate description of ITT analysis was provided in only two RCTs. All trials included an adequate number of participants in the final analysis.
Glycemic control as a value of HbA1c
All six RCTs (17–19, 21–23) provided data on glycemic control described by the HbA1c value at the end of the study. In one study (17), data were available only in graphs, and with no response to our request for crude data, they could not be pooled in the meta-analysis. The pooled results of five trials (n = 136) showed a significant lower HbA1c value in group treated with CSII compared with MDI group (WMD −0.24, 95% CI −0.41 to −0.07, p < 0.001) (Fig. 1). The studies that qualified for analysis were homogenous (χ2 = 3.14, p = 0.53, I2 = 0%). Respective HbA1c values after 3 months are shown in Fig. 2.
Change in HbA1c between baseline and end of the study
Four RCTs (n = 96) provided information on differences in HbA1c values between baseline and the end of the study (17, 19, 21, 22), and one additional study (n = 37) provided crude values (18). Three of these studies (20, 23, 24) found no significant difference between the groups. One RCT (23) found a significant increase in HbA1c value in the CSII group compared with the control group, and one RCT (21) found a significant decrease in HbA1c value in the CSII group compared with the control group (p < 0.02). Because of significant heterogeneity across studies (χ2 = 190.2, I2 = 97.9%), we considered it inappropriate to perform a meta-analysis.
Total insulin dose
The total insulin dose was assessed in three RCTs involving 74 patients (17, 19, 22). Insulin dose values were presented in the unit-per-kilogram-per-day format. A significant lower insulin dose was recorded in the CSII group compared with MDI group (WMD −0.22 IU/kg/d changes, 95% CI −0.31 to −0.14, p < 0.001) (Fig. 3).
BMI and weight
BMI results were provided in two RCTs (18, 22). As BMI was presented as a standard deviation score (SDS) or percentile for age, we could not pool the results for a meta-analysis. DiMeglio et al. (n = 37) did not observed significant difference between groups in BMI percentile for age at the end of the study (18). In contrast, Weintrob et al. (22) found a significantly higher BMI SDS in the group treated with MDI compared with patients treated with CSII (one RCT, n = 23, mean difference 0.35 ± 0.83 vs. 0.37 ± 0.12, p = 0.012).
Severe hypoglycemia and ketoacidosis
Only severe hypoglycemia episodes were assessed in our analysis because of different definitions of mild and moderate hypoglycemia used by the investigators. Moreover, hypoglycemic episodes were established based on continuous subcutaneous glucose monitoring as well as self-monitoring of blood glucose. Wilson et al. (17), DiMeglio et al. (18), Doyle et al. (19) and Weintrob et al. (22) analyses found less episodes of severe hypoglycemia (SH) with CSII than with MDI (4 vs. 10). Two episodes of ketoacidosis were reported by Fox et al. (21) and Doyle et al. (19) in the CSII group, and no DKA episode occurring in MDI group. Although, the difference in rate of SH and DKA (n/patient-year) was not significant (Fig. 4).
Quality of life
QoL, satisfaction with treatment, and psychoemotional condition of the parents were assessed in four RCTs (n = 98) (17, 19, 21, 22) (Table 2). Because of differences in the age and maturity of the study subjects as well as the various forms of diabetes QoL questionnaires used, the results were analyzed separately. In one of the studies, no description of the tool used to assess the QoL was provided. In the study by Weintrob et al. (22), a comparison of results at the end of the study showed a significantly higher level of satisfaction with treatment in patients treated with CSII; this was the only study to provide the total score for both groups at the end of the study (30.6 ± 3.7 vs. 21.9 ± 3.8, p < 0.001), without any significant difference in the impact, worry, or satisfaction subscales. Wilson et al. (17) noticed an improvement in QoL from baseline until the end of the study in the group treated with CSII compared with the MDI group (ΔDQoL −0.24 ± 0.25 vs. −0.08 ± 0.19, p = 0.03), but they did not compare the two groups at the end of the observation. Only Fox et al. (21) assessed QoL separately for the father and the mother. The authors found that the mothers of patients on MDI reported a greater impact of diabetes on the family life than those of patients on pumps. Moreover, the fathers of children in the MDI group reported significantly higher scores on the stress index. The total score was not provided in this study. No difference in the Pediatric Diabetes QoL between the two groups was reported. In Doyle et al. (19) study, the modified for youth the Diabetes QoL questionnaire was implemented, which was filled out by the half of the study population. There was no difference in total scores.
Table 2. The questionnaires used to assess the QoL in children with 1 diabetes mellitus
DCCT, Diabetes Control and Complications Trial; QoL, quality of life.
Diabetes Quality of Life – a modified quality of life measure for youths
Investigators in two RCTs (n = 41) reported that all participants opted for continuation of the pump therapy at the end of the study (17, 21). In the DiMeglio et al. (18) study, 19 of 20 families continued the CSII treatment. Weintrob et al. (22) reported that 16 of 23 patients from both the control and the experimental groups decided to switch to pumps.
The principal conclusion from this meta-analysis is that CSII is more effective than MDI in getting lower value of the HbA1c in children with T1DM. This difference of HbA1c was 0.24%, but it was statistically relevant. Findings were stable with both fixed and random effect models. This effect of lower HbA1c with CSII than with MDI was observed at 3 months of study duration as well as at the end of intervention. Interestingly, only one (19) of the six RCTs analyzed separately showed a significantly lower HbA1c in CSII group compared with MDI group. Other potential benefits of the CSII therapy included a relatively low, compared with the control, insulin requirement without any tendency toward changes in BMI (18, 19, 22). Patients who had received CSII therapy compared with patients on MDI had low rate of Ketoacidosis, severe hypoglycemic events and also had improved QoL. The results of presented meta-analysis is consistent with the consensus statement from the European Society for Paediatric Endocrinology (ESPE), the Lawson Wilkins Pediatric Endocrine Society, and the International Society for Pediatric and Adolescent Diabetes (ISPAD), concerning CSII in children with T1DM presented by Phillip et al. (24). The performed meta-analysis adds category A evidence into this recommendation.
All trials suffered from limited sample sizes. Moreover, methodological limitations included unclear randomization methods and lack of ITT analysis. An additional limitation is the lack of data concerning standard deviations in HbA1c changes; in one study (17), HbA1c values were only presented on a graph and, consequently, had to be excluded from this analysis. Moreover, in spite of direct requests, some authors failed to provide the required information. Because of differences in the presentation of body weight (BMI z-score, BMI SDS, and BMI in percentile for age and gender), we could not meta-analyze the impact of the intervention on growth or development. The lack of a standard QoL questionnaire for the parents of preschool children made it difficult to carry out an objective assessment. Three studies (17, 18, 21) were conducted in young children but with different questionnaires. Furthermore, the measured outcomes were not presented in all studies, making it difficult to analyze the data; this included, for example, poor presentation of data about insulin doses or physical development. Moreover, studies included in this analysis were conducted for no longer than 12 months, most of them were carried out for period of 3 or 6 months. Although, this meta-analysis shows a better metabolic control regarding the value of HbA1c with CSII, it does not allow to predict the impact of CSII method on metabolic control in longer period of treatment.
Implications for practice and research
The findings of this meta-analysis suggest that a QoL questionnaire should be developed for the parents of preschool children. Even though QoL is one of the most important parameters, it has thus far been difficult to assess objectively; this is because no standard QoL questionnaire has been developed for the parents of small children that would take into account the impact of the disease, satisfaction with treatment, or the parents’ anxiety level. Also, larger sample sizes are necessary. The fact that this analysis has shown that individual controlled studies have not demonstrated any statistically significant difference in HbA1c levels between patients receiving CSII and MDI may have been because of the insufficient number of patients enrolled. Only one trial with the participation of adolescent patients was conducted. No trials over a period of more than 12 months have been carried out to assess the impact of treatment not only on metabolic control but also on the incidence of severe hypoglycemic events, the child’s emotional development, the incidence of depression, anorexia, and the ability to take on independent employment in adult life.
The analysis gives light on the methodology of study like a necessity of presentation primary end-points in absolute values with standard deviation. Moreover, it is necessary to standardize physical development parameters, taking into account the dynamics of changes resulting from the child’s development as well as population-related and gender-related differences.
Five of the six RCTs taken separately did not show any improvement in glycemic control in patients treated with pumps compared with MDI. However, this meta-analysis of RCTs indicates that CSII therapy is a more effective form of insulin therapy compared with traditional MDI. In short period of treatment, this method does not increase the risk of DKA. The low rates of discontinuation indicate that pump therapy is preferred by children.