Effectiveness of cognitive–behavioural therapy on glycaemic control and psychological outcomes in adults with diabetes mellitus: a systematic review and meta‐analysis of randomized controlled trials

Diabetes is a chronic progressive condition presenting physical, social and psychological challenges that increase the risk of comorbid mental health problems. Cognitive–behavioural therapy (CBT) is effective in treating a variety of psychological disorders, and may potentially improve glycaemic control and psychological outcomes in diabetes. This systematic review and meta‐analysis aims to establish the effectiveness of CBT on glycaemic control and comorbid diabetes‐related distress, depression, anxiety and quality of life in the short, medium and longer term among adults with diabetes.


Introduction
Diabetes is a chronic medical condition that requires people with diabetes to engage in a lifelong therapeutic selfmanagement regimen in order to maintain glycaemic control [1,2]. The diagnosis of diabetes and efforts towards selfmanagement, particularly lifestyle modification, the demands of a daily treatment regimen and thoughts about the risk of developing diabetes complications, are behaviourally and psychologically challenging [3]. It is estimated that 50% of patients demonstrate decreased psychological states at the time of diabetes diagnosis [4]. Among people with diabetes, commonly observed comorbid mental health conditions include diabetes-related distress, anxiety and depression resulting in poor glycaemic control and reduced quality of life [5][6][7]. Diabetes-related distress was found to affect 13.8-44.6% of people with diabetes [8]. Diabetes doubles the odds of comorbid depression and 12-27% of people with diabetes experience depression at a rate two to three times that of the general population [9][10][11]. Anxiety also occurs iñ 14% of people with diabetes, and 40% show elevated levels of sub-syndromal anxiety [12]. The relationship between diabetes and co-occurring psychiatric disorders is complex and bidirectional because they both influence each other and are affected by biological pathways, social and psychological factors [13].
This review addresses psychological disorders whose pathogenesis results as a complication of living with diabetes.
Correspondence to: C. Uchendu. E-mail: ntxcou@nottingham.ac.uk Given that the barriers to coping with diabetes management are mostly cognitive and behavioural, rather than related to insufficient knowledge or skill, an intervention comprising cognitive and behavioural components may result in improvement. Therefore, this review focuses on cognitivebehavioural therapy (CBT) because it is recommended as the primary psychological therapy for effectively challenging dysfunctional thoughts, beliefs and negative behaviours in people with long-term conditions and replacing these with cognitions that are more self-helping and realistic [14][15][16]. This reduces the feeling of being overwhelmed and aids effective coping with the demands of daily stressors and the treatment regimen.
Given the mixed results on the effect of CBT on adults with diabetes, this review seeks to determine its pooled effectiveness among people with diabetes in the short, medium and longer term.
Previous systematic reviews have examined various psychological interventions in same review and found that CBT did not improve glycaemic control in adults [5,[33][34][35] and have not considered the duration of effects of CBT [36]. Nevertheless, these studies found that CBT improved depression and other psychological outcomes. It is noteworthy that the combination of various psychological interventions in one review might mask the overall effectiveness of each single included intervention. Therefore, this review focuses specifically on studies that performed CBT on adults with Type 1 diabetes, Type 2 diabetes or a combination of both. The justification for this choice is that, although Type 1 and Type 2 diabetes may be physiologically different, the psychological challenges patients face are similar, especially as insulin is increasingly being used in the treatment of Type 2 diabetes. All patients with diabetes are all predisposed to diabetes-related distress, anxiety, depression, suboptimal glycaemic control and reduced quality of life. A study on the duration of the effectiveness of CBT is warranted because it will aid in planning for psychological care in diabetes management. This is the first systematic review and metaanalysis of RCTs to examine the effectiveness of CBT on glycaemic control and comorbid psychological outcomes in the short, medium and longer term in adults with diabetes.

Aim of the review
The aim was to critically appraise, synthesize and systematically review the available published evidence on the effectiveness of CBT on glycaemic control and comorbid diabetes-related distress, depression, anxiety and quality of life in the short, medium and longer term among adults with diabetes.

Research question
In adults with diabetes, what is the effectiveness of CBT on glycaemic control, comorbid diabetes-related distress, depression, anxiety and quality of life in the short, medium and longer term?

Study eligibility
Studies eligible for inclusion in the review included only those with study populations of adults ≥ 18 years clinically diagnosed with either Type 1 or Type 2 diabetes of ≥ 6 months duration. Participants had at least one of: HbA 1c , diabetesrelated distress, anxiety, depression, or quality of life score above normal limit for the standardized scale used.
The study intervention was psychotherapy involving the combination of cognitive and behavioural strategies (CBT), which had to be more than a one-off session. This includes CBT delivered face-to-face, via the telephone or internet, to individuals or groups, and by psychologists or trained therapists. Comparison groups included those who received non-CBT interventions, usual care or waiting list controls. The primary outcome of interest was glycaemic control. Secondary outcomes were diabetes-related distress, anxiety, depression and quality of life. Study designs were only RCTs. All outcomes were measured using validated scales.

Search methods
The search strategy was aimed at locating published and unpublished studies. There was no limitation on geographical area, language or year of publication. Comprehensive search filters were developed for a database search using the identified keywords and index terms across all included databases (Doc. S1). Included databases were: MEDLINE (1946-2014), CINAHL, Web of Knowledge, PsycINFO (1806-2014), Embase (1974-2014), Cochrane Central Register of Controlled Trials and PubMed. Websites such as www.controlled-trials.com, www.clinicaltrials.gov and www.who.int/trialsearch were also searched for ongoing trials. The reference lists of all review articles were searched extensively by hand for additional studies. Two reviewers (CU and HB) independently assessed the abstracts of relevant articles identified from the search. Studies with abstracts that did not conform to the inclusion criteria were excluded. Full articles of abstracts that suggested potential eligibility were retrieved. Consensus was reached through discussion. Studies were included based on agreement by the authors and reasons for exclusion were recorded. Disagreements were presented to a third reviewer for a final decision on the inclusion or exclusion of the article.
In total, 1144 studies were identified from the search; 35 full-text articles were evaluated against the inclusion criteria ª 2016 Diabetes UK and 12 RCTs met the inclusion criteria for the review (Fig. 1). However, only nine studies were included in the meta-analysis because two studies had insufficient data for inclusion [28,29] and one was a three-arm trial [32]; therefore, their results were reported narratively.

Data extraction and study quality
The authors independently extracted data for the study using the Cochrane collaboration data collection form for RCTs and variations in the data extracted were resolved through discussion. Data extracted included general study information and demographic data, as well as characteristics of participants, interventions and outcomes. Missing data were obtained from trial authors where possible.
The methodological quality of articles was assessed independently by the reviewers, based on the quality criteria specified by the Cochrane Collaboration Risk of Bias Tool [37]. Where consensus could not be reached, a third reviewer was consulted to make the final decision. For the purpose of this review, alterations were made to the tool such that blinding of participants and personnel and incomplete outcome data were separated and two more parameters (similarity of baseline characteristics between groups and timing of outcome measurement) were added. Studies for inclusion in the metaanalysis. n = 9 Articles excluded after full text evaluation with reasons n = 23 3= Not RCTs; 2= Non-English language 4= Mixed interventions; 9=RCTs but not CBT intervention; 5=Not CBT for DM Articles remaining after removal of duplicates. n = 766 Additional records identified through other sources n = 9 Studies included in the systematic review but excluded from meta-analyses due to insufficient data n = 3 Potentially relevant published articles identified by searching databases. n = 1135 Included studies consisted of both methodologically sound and less methodologically sound studies. Methodologically sound studies were categorized as those with seven or more positives on the risk of bias summary [22][23][24][25]32]. Given the small number of studies found, no study was excluded on the basis of methodological quality.
Randomization of participants was performed in nine studies [21][22][23][24][25][26][27][28]32]. Davazdahemamy et al. [29] and Amsberg et al. [30] used the term 'randomly allocated', but the randomization process was unclear because there was insufficient information to aid judgement. Henry et al. [31] had high risk of bias because they generated allocation based on the judgement of the primary care physician. Allocation concealment prior to assignment of study participants to either intervention or control groups was performed in six trials [23][24][25][26][27]32]. Three trials had insufficient information to aid judgement on this criterion [22,29,30]. Three other trials [21,28,31] were judged high risk for selection bias because allocation concealment was not ensured. Safren et al. [24] blinded both participants and personnel, whereas four studies [22,28,30,32] blinded study personnel, but were either unclear or high risk regarding participant blinding. Three studies blinded outcome assessors [22,24,32], whereas others were unclear regarding this criterion. All studies except Davazdahemamy et al. [29] reported on loss to follow-up with reasons for dropout. Intention-to-treat (ITT) analysis was conducted in seven of the studies indicating low risk of attrition bias [21][22][23][24][25]28,30]. van Bastelaar et al. [28] reported a high dropout rate, especially in the CBT group. Three studies [26,27,32] had high risk for attrition bias because they did not perform intention-to-treat analysis, so measures were reported for completers only. Two of the studies were unclear about this criterion [29,31]. In eight studies, outcomes were measured in both completers and non-completers [21,22,[24][25][26][27][28]30]. All included studies had intervention and control groups with similar demographic factors, diabetes-related psychological symptoms and baseline scores for study outcomes. They also reported outcome measures periodically. Wherever other biases could not be ascertained, there was limited information to aid judgement.

Measures of treatment effect
In this review, a control group was defined as participants with either Type 1 or Type 2 diabetes who did not participate in the intervention and against whom comparison with the intervention group will be made. RCT referred to studies that include both intervention and control groups with diabetes, whose participants were randomly allocated to either group at the beginning of the study. Calculation of effect size was based on the difference in change of a measurement before and after the intervention between the CBT and control (non-CBT) groups. Effect sizes (ES) were expressed as relative risk for dichotomous data and standardized mean difference for continuous data. Effect sizes were categorized along a continuum of small (ES < 0.20), moderate (0.33 ≤ ES < 0.55) and large (ES > 0.56) [38]. Standardized mean difference was also used as a summary estimate for the overall effect size with 95% confidence interval (95% CI).

Data synthesis
Data from included RCTs were pooled in statistical metaanalysis and synthesized using the Review Manager software (RevMan v. 5.3 Cochrane Collaboration, Oxford, UK). All data were subject to double entry to minimize error. Fixed effects analysis was performed when heterogeneity was < 50%, whereas random effect analysis was performed when heterogeneity was between 50% and 85% [39]. Where statistical pooling was not possible due to insufficient data, the findings were synthesized narratively. Heterogeneity between studies was assessed using chi-square test and Higgins' I 2 test.

Characteristics of interventions
Although all included studies conducted CBT for intervention groups, they varied in terms of format, duration, total number of sessions and professional background of therapists. CBT was conducted on a weekly basis by psychologists, nurses and/or dieticians who were pre-trained in performing CBT. The nurses worked with psychologists in three studies [26,27,30]. Interventions for CBT groups consisted of cognitive and behavioural components, whereas intervention for the control groups differed slightly across the studies, but was most commonly usual care. The total number of CBT sessions ranged between 6 and 21 sessions, and the duration of sessions ranged between 30 min and 2 h per session, spanning a period of 6 weeks to 4 months. Three studies [21,23,24] administered booster CBT sessions to participants in the intervention groups. Booster sessions refer to sessions that were aimed at reinforcing lessons learnt from ª 2016 Diabetes UK previous CBT sessions. With the exception of van Bastelaar et al. [28], who performed web-based CBT, all studies were carried out within healthcare settings. The format of CBT across studies included face-to-face individual sessions [24,25], face-to-face group sessions [21,22,26,27,[30][31][32], individual telephone sessions [23] and web-based sessions [28]; one study did not specify the format used [29].

Characteristics of participants
The total number of participants in the included studies was 1445, and a total of 354 participants dropped out of the studies. The highest dropout rate was recorded in van Bastelaar et al. [28] and the lowest in Henry et al. [31]. The mean age of participants was between 37.4 AE 11.1 and 61 AE 10.8 years. The mean duration of diabetes was between 6.4 AE 8.7 and 21.6 AE 10.8 years. Two trials did not provide data on participant's mean duration of diabetes, although participants ought to have had diabetes for ≥ 6 months to be eligible for recruitment into the study [23,29].
Inclusion criteria were similar across all studies: age ≥ 18 years; ability to read and write; diabetes duration of ≥ 6 months; HbA 1c levels, diabetes-related distress, anxiety, depression or quality of life score above specified cut-off points for validated scales used. The most commonly cited exclusion criteria reported in the included studies were pregnancy, severe medical comorbidity, and the presence of current alcohol/substance abuse disorder, bipolar depression or any other psychotic disorder. All study outcomes were measured using validated scales. Diabetes-related distress was commonly measured with the Problem Areas in Diabetes (PAID) scale. Depression was measured using different tools, including the Beck Depression Inventory (BDI; ≥ 14), the Centre for Epidemiological Studies Depression Scale (CES-D; ≥ 16) and the Hospital Anxiety and Depression Scale (HADS; 1-7). Anxiety was commonly measured using the State-Trait Anxiety Scale (≥ 35), although the HADS (1-7) and Depression Anxiety Stress Scale (DASS-42) were also used.

Characteristics of study outcomes
All studies measured outcomes at baseline and follow-up periods with some trials providing follow-up data at specific intervals up to 12 months post intervention.
Of the studies that could not be pooled in the metaanalysis, one trial reported that CBT had a significant effect in decreasing HbA 1c score (P = 0.005) [29]. Conversely, another found no significant short-term treatment effect of CBT (P > 0.05) on HbA 1c even when only participants with elevated baseline HbA 1c levels were examined [28]. One study performed a three-arm trial of CBT, attention control and individual control groups, but found that although all three groups showed improved HbA 1c levels after 3 months, the CBT groups showed more improvement than other control groups [mean HbA 1c change at 3 months: CBT group À32 mmol/mol (À0.8%) vs. À28 mmol/mol (À0.4%) for attention group controls and À28 mmol/mol (À0.4%) for individual controls] [32].

Diabetes-related distress
Five studies examined diabetes-related distress [26][27][28]30,32]. Because of the varying effect measures reported across these studies, their results were not pooled in a metaanalysis. Van der Ven et al. [26] and Van Bastelaar et al. [28] reported a short-term decrease in diabetes-related distress (1 month, P = 0.01 and 3 months, P < 0.001, respectively), whereas in Weinger et al. [32], CBT did not improve diabetes-related distress at short-, medium-and long-term measurements. In Amsberg et al. [30], a significant effect was observed between CBT and control groups regarding diabetes-related distress in the medium term (À6.88; 95% CI À11.50 to À2.25; P = 0.019), which improved at 48-week follow-up (P = 0.004). In Snoek et al. [27], there was no significant difference in diabetes-related distress score between the CBT and control groups. Even at 6-month and 1-year follow-up, participants still reported high levels of diabetes-related distress (P = 0.99 and 0.68, respectively).

Depression
Eleven of 12 included RCTs measured depression at baseline [21][22][23][24][25][26][27][28][29][30][31], eight of which had sufficient data to be included in the meta-analysis. The forest plots present short-term (up to 4 months), medium-term (up to 8 months) and long-term (up to 12 months) results for depression. Given that different psychometric scales were also used for assessing depression, the standardized mean difference was used as a summary statistic in order to standardize the results of the studies to a uniform scale.
Four studies were analysed for the short-term effect of CBT on depression [21,24,26,31]. With a heterogeneity level of 23%, the fixed effect method analysis found a large effect size and a significant difference in depression scores in favour of CBT (À0.52; 95% CI À0.79 to À0.26, P = 0.0001). This indicates a significant short-term improvement in depression in favour of CBT (Fig. 3a). The findings from trials by Davazdahemamy et al. [29] that were not included in the meta-analysis also revealed that CBT had positive short-term effect on depression with Pvalues of 0.012 and < 0.001, respectively. Five studies [21,24,25,27,30] provided information for a medium-term effect of CBT on depression. With a substantial heterogeneity level of 73%, the random effect method was used for data synthesis. The resulting finding was a moderate effect size (À0.43; 95% CI À0.79 to À0.06, P = 0.02). This result indicates that CBT had a significant medium-term effect on depression (Fig. 3b). Interestingly, on exclusion of the studies that did not provide booster sessions, a non-significant result was found with a large overall effect estimate of À0.54 (95% CI À1.28 to 0.21, P = 0.16). The five trials [23][24][25]27,30] that measured the long term-effect of CBT on depression were included in a meta-analysis (Fig. 3c). Owing to a moderate heterogeneity level of 44%, the fixed effect method was used to synthesize the data. The result of the analysis demonstrates a significant difference in depression scores with a further reduced and fairly small effect size at 12 months follow-up in favour of CBT (À0.26; 95% CI À0.41 to À0.10, P = 0.001). On exclusion of studies that did not offer booster sessions to participants, the I 2 value decreased to 0%, hence the fixed effect model analysis found a significant effect (P = 0.003). Also, an increased but moderate effect size was found for the long-term effect of CBT on depression (À0.38; 95% CI À0.59 to À0.17) from studies that offered booster sessions.
The study by Lustman et al. [22] provided dichotomous data on depression outcome with regards to remission of depression and clinically significant improvement in depressive symptoms. For each result, data were provided at 3-(post intervention) and 6-month follow-up, respectively, and produced a trend towards significance, because CBT participants were 2.31 times more likely to achieve remission of depression than controls risk ratio (RR) 2.31; 95% CI 1.45 to 3.67, P = 0.0004). From the results, it is also evident that CBT participants were 1.88 times more likely to achieve a clinically significant improvement in depressive symptoms compared with controls (RR 1.88; 95% CI 1.25 to 2.84; P = 0.003). The test for overall effect also suggested a significance difference (P < 0.00001) between CBT participants and controls. However, these findings are the result of a single RCT and should, therefore, be interpreted with caution. Four trials [21,[29][30][31] examined of the effect of CBT on anxiety, only three of which were included in the metaanalysis due to lack of sufficient data in one of the trials [29]. This analysis was completed according to periods of anxiety outcome measurement (3, 6 and 12 months). Penckofer et al. [21] reported both state anxiety and trait anxiety and these were included in the meta-analysis as two separate anxiety results. The fixed effect method was used in pooling the data because the level of heterogeneity was 32% (Fig. 4). The results yielded a large effect size and indicated that there was a significant reduction in mean anxiety between groups in favour of CBT compared with the control groups at 3 months (À0.55; 95% CI À0.88 to À0.21, P = 0.001). However, one of these trials [31] produced a greater reduction in anxiety than observed in the other studies although this may be related to the smaller sample size (n = 19). Nevertheless, on exclusion of this study, there remained a significant improvement in anxiety with a moderate effect size (À0.41; 95% CI À0.76 to À0.06, P = 0.02) in the CBT groups compared with the control groups and heterogeneity reduced to 0%. At 6 months, although the overall effect estimate did not increase significantly, there was still a significant difference in anxiety scores in favour of the CBT groups (À0.56; 95% CI À0.85 to À0.27, P = 0.0002). At 12 months, only one trial [30] measured anxiety and found no significant difference in anxiety level between CBT and control groups (À0.33; 95% CI À0.79 to 0.13, P = 0.16). However, the test for overall effect yielded a large effect size and a significant improvement in anxiety levels (À0.51; 95% CI À0.71 to À0.31, P < 0.00001). This finding from just one study should be interpreted with caution.

Quality of life
Four trials examined the effect of CBT on quality of life [21,25,32,40]. Given the varying scales used, their results were presented in dissimilar patterns and unsuitable for synthesis in a meta-analysis. Penckofer et al. [21] reported improvement quality of life (P < 0.001), especially on the mental and psychological components compared with the control group in the short and medium term. In Davazdahemamy et al. [29], CBT also effectively improved shortterm quality of life (P = 0.011). However, little improvement was seen in Welschen et al. [25] between the CBT and control groups. The CBT group showed a statistically significant improvement in quality of life between baseline and 6 months, whereas participants in the control group showed deterioration in their quality of life. Also, Weinger et al. [32] found significant improvement in quality of life for short-term follow-up although this was not sustained in the medium and long term, respectively, because no significant improvement was observed at 6-and 12-month follow-up.

Glycaemic control
Concerning meta-analysis of studies that measured shortand medium-term glycaemic control, there was a significant short-term reduction in mean HbA 1c in favour of CBT when compared with control groups and a significantly greater reduction was evident in the medium term. This indicates that CBT interventions may hold possible benefits for these time points. The findings support the view of Turner et al. [41] that CBT holds potential clinically relevant benefits for people with suboptimal pre-treatment levels of HbA 1c . The improvement in glycaemic control could be attributed to the effect of CBT in changing negative thoughts, attitudes and beliefs regarding diabetes, which is likely to lead to a change in diabetes self-care behaviours and subsequent glycaemic control [42]. However, the effect of CBT on glycaemic control was not maintained on the longer term, even when booster CBT sessions which have been shown previously to sustain CBT outcomes were provided [24,43]. This might be due to the generally acknowledged difficulty in achieving behaviour change, thus people have the tendency to return to their usual habits due to an inablility to incoporate new bahaviours into their daily lives [25]. It is possible that the CBT interventions were too short to generate lasting effects, or were not tailored to the specific needs of people with diabetes. However, there was insufficient evidence to draw conclusions about long-term effects due to the small number of studies that included long-term outcomes of CBT. Diabetes is a lifelong condition requiring continuous psychological assessment and management in order to stabilize peoples' behaviours, which, in turn, translates into significant improvement in glycaemic control. Further well-designed RCTs are required to demonstrate whether CBT (with or without booster sessions) can sustain glycaemic control in the long term. These findings are important because they represent the overall treatment effect from studies recruiting people with suboptimal HbA 1c .

Effect of CBT on psychological outcomes
CBT significantly improved depression in adults with diabetes in the short, medium and long term, although the effect sizes reduced over time. The positive effect of CBT on depression in this review is congruent with results reported for the effectiveness of CBT in treating depression in those with conditions such as chronic pain, stroke, social phobia, post-traumatic stress disorder and HIV/AIDS. CBT has even been reported to be somewhat superior and more tolerated than antidepressants in treating adult depression [44,45]. Because depression is characterized by repetitive and uncontrolled negative thoughts that generate feelings such as guilt, low mood and low self-esteem [46], these positive effects could be attributed to the influence of CBT in challenging and replacing dysfunctional thoughts with positive and life-enhancing thoughts. The high dropout rate in some studies that measured depressive symptoms might be due low self-esteem, lack of motivation and pessimism associated with depression [47]. This review did not compare between trials with depressed and non-depressed people with diabetes, therefore, the results reflect that of people with diabetes and comorbid depression.
Although CBT seemed to be effective in the short-term management of diabetes-related distress, results at the medium and long term were mixed. It should be noted that the included studies measuring diabetes-related distress also measured depression. Where depression improved and the improvements in depression (although reducing in magnitude) were sustained over time, there was non-improvement in diabetes-related distress over time. This may potentially be due to a lack of tailoring of the CBT interventions to the problem areas specific to diabetes self-care. Recent research emphasizes the importance of distinguishing between diabetes-related distress and depression. Fisher et al. [48] suggests that among people with diabetes, depression and diabetes-related distress may present with similar symptoms, but are not necessarily the same. The latter develops from living with diabetes, therefore, interventions need to be tailored to the specific needs of the different groups The mixed findings for diabetes-related distress warrant further research with specifically designed CBT interventions to highlight distinguishing factors for diabetes participants.
Overall, CBT significantly reduced anxiety levels in the short and medium term, and potentially in the longer term. This suggests that CBT may be an effective treatment for anxiety among adults with diabetes. Because of the limited number of studies examining anxiety levels as an outcome, the findings should be interpreted with caution. Notwithstanding, this finding is not surprising because CBT has been noted to be effective in the short-, medium-and longer-term anxiety management, as well as superior to pharmacotherapy for anxiety disorders [45]. The finding of this review is congruent with previous reviews demonstrating the effectiveness of CBT on anxiety in adults within the general population [49,50]. The significant reduction in anxiety levels in participants might be attributed to the acquisition of problem-solving skills that enabled participants to modify negative thoughts and engage in behaviours and/or activities needed to cope adequately with diabetes. Also, participants in these studies were provided with information about diabetes as part of the CBT intervention, which may have further helped to allay condition-related anxieties and aid coping.
Poor diabetes management and psychological health negatively impact quality of life. Although there is growing interest in both quality of life outcomes and features in the therapeutic goals of diabetes care, they are rarely the focus of most interventions. This perhaps explains the paucity of studies measuring the effect of CBT on quality of life. Based on limited evidence, this review showed mixed results for quality of life in people with diabetes, with CBT improving quality of life in some studies and having non-significant effect in others, especially in the medium and long term. Nevertheless, the positive findings concur with those of Welschen et al. [51] who reported that, in diabetes, the relationship between CBT and quality of life is indirect such that improvements in psychological status and glycaemic control can result in concomitant improvements in quality of life.

Strengths and limitations of the review
This study included only RCTs, which are the gold standard of research on the effectiveness of interventions and which yield the highest quality of evidence [52]. Bias was minimized by using the Cochrane Risk of Bias Tool, and critical appraisal of potentially eligible studies by at least two reviewers. Meta-analysis was used to generate the most findings of the review. With regards risk of publication bias across included studies, Duval's trim and fill method, which aims both to identify and correct for funnel plot asymmetry arising from publication bias [53], showed negligible effect size shifts with included imputed studies. Furthermore, using the Begg's funnel plot, no difference between observed and imputed studies was shown. This indicated that there is little evidence of publication bias and that the reported effect is valid.
It should be noted that this review was not a priori registered and did not have an a priori review protocol. Although the review methodology was rigorous, there is a possibility that we failed to include some published or unpublished studies. Owing to limited resources, studies published in foreign languages could not be translated or included. The overall quality of the included RCTs was not optimal and none were conducted in the United Kingdom, Africa or Asia. Most studies neither provided follow-up data nor performed intention-to-treat analysis despite high attrition rates. Furthermore, due to a lack of relevant trials, this review does not provide information on cost-effectiveness, gender-specific effects and the length or format of CBT sessions that are most beneficial and preferred by people, or whether people differing in type of diabetes may benefit more or less from CBT interventions. Subgroup analysis between Type 1 and Type 2 diabetes was not undertaken because this was beyond the scope of this review. Nevertheless, statistically significant improvements in the magnitude reported in this review demonstrate the potential positive effects of CBT on glycaemic control and psychological outcomes. However, regarding the generalizability of our findings, some caution is required because we were unable draw a robust conclusion due to the small number of trials that met the inclusion criteria, the majority of which had small sample sizes. There is a need for further research that examines whether the positive effects of CBT are sustained over several years, to confirm the clinical importance of this intervention in the management of diabetes as a life-long condition.
Agreement and disagreement with previous reviews As reported previously, few reviews have examined the effects of non-pharmacological psychological interventions on glycaemic control. Only one review specifically examined CBT for glycaemic control in diabetes [36], but found no overall statistically significant impact from meta-analysis of CBT on glycaemic control. Nevertheless, only four trials were included in the meta-analysis and durations of the expected effectiveness of CBT were not highlighted. In recognition of the limited human and financial resources in health services in various countries, it is necessary to ascertain the supposed duration of effectiveness of any intervention for planning purposes.

Recommendations for further research
Well-designed RCTs with larger sample sizes are required to assess the effectiveness of CBT in improving glycaemic control over the longer term. More studies are needed that assess the impact of CBT over time on condition-specific psychological outcomes, such as diabetes-related distress. There is scope to increase knowledge about the best mechanisms for the delivery of CBT in this group. Although a group CBT format has clear advantages in terms of efficiency of delivery, it is not yet clear which CBT format (group, individual, face-to-face, web-based) is preferred by people with diabetes, and we need to better understand the stage of diabetes in which delivery of CBT is most effective. Future research might examine potential gender-specific effects of CBT intervention, the effectiveness of CBT in preventing poor glycaemic control amongst people with normal glycaemic control, and barriers to the long-term effectiveness of CBT in those who do not improve. Qualitative research should be undertaken on participant experiences and satisfaction with CBT intervention. With advances in information technology and the rapid increase in web-based interventions for selfmanagement of chronic illness, trials of web-based CBT treatments in other conditions have reported patient outcomes that are comparable with those found with face-to-face interventions [54][55][56]. There is certainly scope to further investigate the impact of web-based CBT intervention specifically for adults with diabetes. In the included studies, the trained personnel delivering the CBT were most commonly diabetes nurses, and this may have been due to the cost implications of CBT delivered by psychologists or trained health professionals from services outside the clinical team. However, the included studies did not provide sufficient ª 2016 Diabetes UK information to examine the cost implications of delivering CBT to people with diabetes, which needs to be assessed. Finally, this review revealed a lack of research evidence on CBT and diabetes conducted in the UK, Africa and Asia. There is a need for well-designed RCTs in these settings to enhance the generalizability of findings.

Conclusions and implications for practice
In people with Type 1 and Type 2 diabetes, CBT is effective in improving short-and medium-term glycaemic control. No significant effect was found for long-term glycaemic control, although there are limited studies measuring longer term clinical outcomes. CBT improves depression in people with diabetes in the short, medium and longer term. It improves anxiety in the short and medium term, and has potential for longer term benefits although there are too few studies assessing long-term outcomes to draw firm conclusions. The evidence for the effects of CBT on diabetes-related distress and quality of life is mixed, and based on a small number of studies with heterogeneity in interventions and reporting of outcomes.
We conclude that CBT-based approaches may be beneficial in diabetes care, although it is recommended that programmes to ensure the maintenance and sustainability of its positive effects are incorporated into diabetes management. Psychologists should work together with diabetes care teams to promote psychological care for people with diabetes, and there is a need for diabetes care teams to be specially trained in current approaches for CBT to enhance its utilization as part of clinical practice. This review recommends the provision of CBT for adults with diabetes to improve psychological health and instil coping strategies for better self-management, which will invariably enhance clinical outcomes for diabetes management and impact on quality of life.

Funding sources
None.