Aligning with the conceptual model applied to our previous review (12), the framework by Chodosh et al. (13) was used to describe self-management programs through five categories: tailoring; group setting; feedback; psychological emphasis; and medical care. Studies were characterized using the five elements based on the most dominant characteristic with the assumption that the outcomes are associated with that study characteristic.

Relevant biomedical databases, including Medline, EMBASE, PsychINFO, CINAHL, and Cochrane Database of Randomized Controlled Trials were searched from November 2013 to July 2023 by applying the previous search strategy developed in consultation with a medical research librarian (12). Please refer to Supplementary Table 1 for an overview of our search strategy. References from relevant guidelines, systematic reviews, and meta-analyses were also examined for relevant publications. The PRISMA checklist for systematic reviews was followed throughout the completion of our review ( Supplementary Table 2 ).

The eligibility criteria from our previous review were applied to assess all citations (12). Briefly, the inclusion criteria consisted of peer-reviewed literature, written in English, and (1) was a randomized, controlled trial; (2) included a population of older adults ≥ 65 years of age that live with Type 2 diabetes; and (3) report the differential effect of self-management support strategy or education on glycated hemoglobin A1C (HbA1c), metabolic outcomes (fasting blood glucose, body mass index [BMI], weight, lipids, blood pressure), or patient-reported outcomes (quality of life, behaviour change, diabetes-related distress, diabetes knowledge, self-efficacy). Publications were excluded if: (1) data was reported on subjects without diabetes, < 65 years of age; (2) the self-management education intervention included testing new or combination oral anti-diabetic agents; (3) no statement of informed consent being obtained was included.

Three research team members (PA, MJ, SK) reviewed citations independently at the title and abstract and full-text levels. Quality assessment and data extraction were completed by three independent team members (PA, DFL, MJ). Disagreements at the full text, data extraction, and quality assessment levels were resolved through discussion between team members. Data extraction was completed using standardized forms and included information on objectives, demographics, self-management interventions, and outcomes for each publication. Quality appraisal was completed using Cochrane’s risk of bias tool (14), and studies were ranked based on having a low, unclear, or high risk of bias.

All data analyses were planned a priori. A meta-analysis was used to combine the results across studies by outcome using the published continuous data from included studies. Specifically, we used the change from baseline to immediate post-treatment or follow-up data (mean change score, standard deviation of mean change score) for both intervention and control groups. In the studies where a measure of variance was reported as confidence intervals, standard error or p-values, we used Cochrane-recommended methods to convert to standard deviation (15). We used a multi-level meta-analytical approach (where applicable) to account for statistical dependence i.e., dependency in effect sizes introduced either by comparison of multiple intervention arms within a study to a common control group or by multiple outcome measures or sub-outcome measures of a primary outcome of interest within a study (such as various measures of patient-reported outcomes) (16–18). The statistical heterogeneity I² statistic was also estimated in the context of a multi-level meta-analytical approach i.e., within-cluster heterogeneity (multiple arms from the same study) and between-cluster heterogeneity (effect sizes across studies). Overall I² for each summary effect size was estimated to represent the heterogeneity not attributable to sample error and is the sum of within-cluster and between-cluster heterogeneity (19).

The summary measures of effect were generated as mean difference (MD) and standardized mean difference (SMD). The units of measurement for clinical outcomes such as weight, lipid profile measures, and blood glucose were converted into the same standard units (mmol/L for lipid outcomes and glucose, and kg for weight) before input into metanalysis. The SMD was used as a summary statistic for PROMs because the studies in this systematic review often assessed the same outcome using different outcome measures or tools. Given this, it was necessary to standardize the results of the studies before they could be compared across studies or combined in a quantitative synthesis. The SMD is interpreted based on its magnitude according to Cohen d recommended thresholds (~0.2=small effect, ~0.5=medium effect, ~0.8=large effect) (20, 21).

Cochran’s Q (α=0.05) was employed to detect statistical heterogeneity and I² statistic to quantify the magnitude of statistical heterogeneity between studies where I² >50% represents moderate and I² >75% represents substantial heterogeneity across studies (22). To ascertain statistical stability and robustness of results, further meta-regression analysis was carried out based on pre-specified subgroups of interest where possible based on the evidence available i.e., the focus of intervention based on Chodosh framework (i.e., Tailoring, group, medical, feedback, and psychological emphasis) and intervention length of follow-up (< 6 months, 6 months or more). Publication bias was assessed using funnel plots where there were at least 10 studies in the meta-analysis. All analyses were performed using R software (metafor and dmetar packages) (23, 24).

Our search strategy ( Figure 1 ) identified 7391 title and abstract citations for review, with 328 citations reviewed in full text. A total of 17 studies with 21 unique comparison arms were identified that met our inclusion criteria ( Figure 1 ). Our included studies had a total sample of 5976 older adults (intervention group: n=3510; usual care group: n=2466). Each intervention arm comprised a self-management approach that fit into the five categories of Chodosh’s framework: 1 study - medical care (25); 3 studies - psychological emphasis (26–28); 4 studies – tailoring (29–32); 4 studies - group setting (33–36); and 5 studies emphasized feedback (37–41). The characteristics of the included studies may be found in Supplementary Table 3 . The risk of bias was assessed for each study and determined that all of the included studies were considered as unclear for risk of bias (25, 26, 28–41), except for one study which was considered at high risk of bias (27) ( Table 1 ).

Overall, the pooled effect of diabetes self-management program interventions for older adults resulted in a statistically significant decrease in HbA1C by 0.32% (95% CI -0.44, -0.19; p<0.00) (25–41) ( Figure 2 ). The most effective self-management program intervention on HbA1C was the use of feedback (-0.52%; 95% CI -0.68, -0.36; p<0.01), followed by psychological emphasis (-0.31%; 95% CI -0.45, -0.16; p<0.01); group setting (-0.28%; 95% CI -0.47, -0.09; p<0.01) and tailoring (-0.13%; 95% CI -0.22, -0.13; p=0.01). Self-management program interventions that emphasized medical care were not statistically significant (-0.09%; 95% CI -0.25, 0.08; p=0.03). Self-management interventions that were less than six months in duration resulted in a greater decrease in HbA1C (-0.40%; CI -0.58%, -0.23%; p<0.01) than programs that were longer than six months (-0.24%; 95% CI -0.40, -0.7%; p<0.01).

The pooled treatment effect of diabetes self-management program interventions in older adults on fasting blood glucose (mmol/L) was a reduction in 0.67 mmol/L (95% CI -0.96, -0.39; p<0.00) ( Figure 3 ) with tailored focused interventions resulting in the greatest decrease (-0.86%; 95% CI -1.35, -0.37; p<0.01) (27, 28, 31, 39). Self-management interventions that were less than six months in duration resulted in greater improvements in fasting blood glucose (-0.72 mmol/L; 95% CI -1.09, -0.36; p<0.01) than programs longer than six months (-0.62 mmol/L; 95% CI -1.05, -0.19; p<0.01).

The pooled treatment effect of diabetes self-management program interventions in older adults on BMI was a reduction of 0.31 (95% CI -0.51, -0.06; p<0.01) ( Figure 4 ) (25, 27, 28, 31, 34, 35, 38, 39). The self-management program intervention strategy that was most effective in improving BMI was psychological emphasis (-0.53; 95% CI -0.65, -0.41; p<0.01). Similarly, self-management program interventions that were less than six months in duration resulted in greater improvement in BMI (-0.54; 95% CI -0.65, -0.42; p<0.01) versus programs that were longer than six months (-0.05; 95% CI -0.19, 0.09; p=0.46). Similarly, small weight changes were observed in older adults exposed to self-management program interventions, with a decrease in -1.38 kg (95% CI -1.96, -0.80; p<0.00) ( Figure 5 ) (26, 28, 31, 35, 38). The self-management program intervention strategy that yielded the most weight loss was psychological emphasis (-1.49 kg; 95% CI -1.82, -1.16; p<0.01). In accordance with BMI changes, shorter self-management programs (< 6 months) were more effective in weight loss (-1.42 kg; 95% CI -1.76, -1.08; p<0.01) versus programs longer than six months (-1.15 kg; 95% CI -1.81, -0.48; p<0.01).

The pooled treatment effect of diabetes self-management program interventions in older adults on systolic blood pressure did not result in a statistically significant reduction (-2.20 mmHg; 95% CI-5.49, 1.08; p=0.19), regardless of the length of duration of the program (25, 26, 28, 30, 32, 34, 38, 39). Similarly, although the pooled treatment effect of diabetes self-management program interventions in older adults on diastolic blood was statistically significant (-1.48 mm/Hg; 95% CI -2.73, -0.24; p=0.02) (25, 26, 28, 30, 32, 34, 38, 39), the reduction is clinically negligible.

The pooled treatment effects of diabetes self-management program interventions in older adults on lipids were varied. There was a small but statistically significant reduction in total cholesterol levels (-0.13 mmol/L; 95% CI -0.20, -0.05; p<0.00) (25–27, 31, 32, 38, 39), with tailoring of self-management program interventions as the most effective in reducing total cholesterol (-0.18 mmol/L; 95% CI -0.24, -0.13; p<0.01). Similarly, there was a small but statistically significant improvement in triglycerides (-0.12 mmol/L; 95% CI -0.19, -0.05; p<0.00) (25, 28, 31, 38, 39), with interventions lasting less than six months (-0.16 mmol/L; 95% CI -0.27, -0.04; p<0.01) and offering psychological emphasis yielding the greatest improvement (-0.16 mmol/L; 95% CI -0.29, -0.04; p<0.01).

The pooled effect of diabetes self-management program interventions for older adults resulted in statistically significant improvements in the standardized mean differences in patient-reported outcomes measures for behaviour change, diabetes knowledge and self-efficacy (-0.91 SMD; 95% CI -042, 1.41; p<0.00) (27, 29, 30, 33, 35–38, 40) ( Figure 6 ). The most effective self-management program intervention strategy associated with the greatest improvements in patient-reported outcomes was feedback (1.64; 95% CI 1.08, 2.21; p<0.01). Self-management program interventions that were less than six months were more effective (1.06; 95% CI 0.44, 1.69; p<0.01) versus programs that were longer than six months (0.60; 95% CI -0.26, 1.47; p=0.17).

The updated literature reflected all of the possible categories of Chodosh’s framework for self-management programs for older adults, highlighting an evolution and distribution of self-management approaches (13). The literature also reflected a broader reflection of outcomes, including patient-centred outcomes, such as behaviour change. Of note, many outcomes found some small but significant improvements that are noteworthy as self-management programs are often in conjunction with pharmacotherapy and comorbid conditions that require similar behaviors management, nutrition, and physical activity approaches.

Our updated review included peer-reviewed literature published in English from November 2013 to July 2023. We recognize that literature may have been missed; however, we were able to apply the same search strategy to five databases yielding a large pool of citations. Unfortunately, concerning the representation of literature within the categories of Chodosh’s framework, medical care in self-management programs continues to be minimally represented in the literature to date. We found the literature to be susceptible to bias as most of the studies were deemed to have an unclear risk of bias due to a lack of reporting (14).

Our review builds on the findings of our previous review (12) as it provides valuable insight into new literature that has emerged on this topic, specifically to understand the impact of specific self-management programs on clinical and patient-reported outcomes for older adults living with diabetes. The findings help to better understand the growing body of literature regarding older adults receiving self-management education and support to meet their complex social, cognitive, medical, functional, and psychological needs (2–7). The findings of this review highlight the need for customization of self-management programs to optimize clinical and patient-centred outcomes, specifically noting that feedback strategies and programs of shorter duration (< six months) were most effective. Future research should explore the need for combination strategies (e.g. feedback and psychological emphasis) to address the complex needs of older adults living with diabetes and multimorbidity. Additionally, future research may also consider cost analyses of self-management programs in supporting older adults to remain out of the hospital, decrease the risks and impact of complications, and age in place. Just as no two people are the same, the findings of this updated review demonstrate that to improve different outcomes, customized self-management strategies for older adults should be employed.