Overall, in this pooled analysis, people with diabetes and depression had a 76% increased risk of LEA compared with people with diabetes without depression; this result was statistically significant. There was, however, a high level of heterogeneity between studies (I2=87%). To investigate this further, we conducted various subgroup analyses. Sample size did not explain our findings (with large and small studies both showing increased but insignificant findings) or the heterogeneity. Geographical variation in LEA rates has been previously documented.29 30 Potential reasons why results might differ between continents would include different population characteristics and healthcare systems.31 In this review, the region in which the studies were conducted (USA vs Europe) also yielded increased but insignificant results, and heterogeneity remained. The subgroup analysis by study quality showed that the studies with the least bias (minimal bias) produced a 34% increased risk of LEA in people with diabetes and depression, which was statistically significant.
It must be acknowledged that only five studies were eligible for inclusion in the meta-analysis, and this limits the robustness of the subgroup analyses performed. Thus, while there is an overall increased risk of LEA in people with depression and diabetes, further research is needed including population-based registry data and more methodologically robust methods of recording depression, diabetes and LEA (ie, ICD codes or medical diagnoses).
Strengths and limitations of this review
This is the first systematic review that the authors are aware of that investigates the association of depression with LEA in people with diabetes. A comprehensive and systematic literature search examined six databases, yielding seven studies eligible for inclusion in the systematic review. Although a significant amount of heterogeneity was found between studies in the meta-analysis, this was examined using appropriate statistical measures as well as a priori defined subgroup analyses. The meta-analysis only included studies that reported adjusted estimates, as these would be considered less biased/confounded results, in order to more accurately represent the true effect of depression and risk of LEA.
The systematic review and meta-analysis is not without limitations, however. Caution needs to be applied when interpreting the results of the studies on depression prevalence as many included poor methodological practices such as small sample size and no confirmation of difference between type 1 and type 2 diabetes, and therefore a subgroup analysis by type of diabetes was not possible. There are fundamental differences between the conditions in relation to mean age of onset and the temporal relationship that could be explored in later systematic reviews as additional studies are conducted. In addition, all of the studies were conducted in high-income, developed countries such as the USA, UK and Australia, the results of which would not be generalisable to resource-poor settings. Differences in diet, lifestyle and culture between these countries may also play a role in diabetes care and thus impact the findings. One study included male war veterans only, and these findings would be very specific to this population. There was also significant variation in the number of potential confounders adjusted that may influence the plausibility of the different study findings.
The diagnosis of depression varied according to study and included different scales, self-reported diagnosis and ICD-9 classifications of depression. Self-reported depression is not a reliable method for obtaining true prevalence of this disorder, particularly among men. This could result in either an overestimate or underestimate of the association with LEA. Variability in the exposure definition may explain the heterogeneity in the meta-analysis
Also, depression status may change over time and measurement at one time point only in these studies is a limitation. As is a problem with most observational studies, unmeasured confounding cannot be ruled out.
Of note, Salmi et al reported on 229 956 patients from a national register-based cohort in Sweden, but only an abstract for this work was available. The abstract does not provide detailed information on the definition of depression and LEA, and efforts to contact the authors for clarity proved futile. We have inferred the diagnoses were made using ICD-9 codes (as this is the method used generally in previous research using the Swedish national data). This is an acknowledged limitation of the review, however.
Heterogeneity
There was a great degree of heterogeneity in the current meta-analysis and would be deemed ‘high’ according to the Cochrane criteria for I2 (>75%).21 The authors tried to control for this heterogeneity by using the random-effects model. Under the random-effects model, we allow that the true effect could vary from study to study. For example, the effect size might be a little higher if the subjects are older, or more educated, or healthier; or if the study used a slightly more intensive or longer variant of the intervention; or if the effect was measured more reliably. In addition, we quantified heterogeneity by using the I2 statistic that focuses attention on the effect of any heterogeneity in the meta-analysis, the percentage of total variation across studies is due to heterogeneity and most importantly in the case of the current meta-analysis, the I2 value does not depend on the number of studies in the meta-analysis.18 Ideally, to further explore reasons for heterogeneity, authors would conduct a priori defined subgroup and sensitivity analyses as well as meta-regression using different covariates. These are largely dependent on the number of studies in the meta-analysis, however, and were not feasible for the current meta-analysis.
Recommendations for future research
Differing definitions and assessment tools for diagnosing depression were used in the studies included in this systematic review. The reported prevalence of depression differs between self-report versus clinical interview diagnosis.32 The use of a standard assessment tool for depression would improve comparability of results from different settings. Further research is required to explore whether the level of severity of depression is predictive of complications. Potential confounders needing consideration in future research include age, gender, education, marital status/living alone, socioeconomic status, type of diabetes, duration of diabetes, insulin use,diabetes treatment/control, self-reported health status, risk behaviours, quality of life, foot care behavior, mood, smoking status, alcohol problems, macrovascular complications, microvascular complications, foot-specific complications, severity of complications, other medical or mental health conditions, and healthcare utilisation.
The literature review confirmed clinical practices vary per individual practitioner, per location and per patient.3 The evidence of under-diagnosing depression in patients with diabetes needs to be considered as a potential bias. Such bias would cause over-reporting of LEAs in patients with no comorbidity and under-reporting of LEAs in patients with comorbid depression.