The aim of this study was to conduct a systematic review evaluating the efficacy of interventions aimed at improving medication adherence for chronic diseases in India. The study protocol was registered on PROSPERO (Registration ID: CRD42022345636). A preliminary scoping search was performed to refine search criteria and identify outcome measures. This study was undertaken according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework (30). A PRISMA (2020) checklist is shown in the Supplementary material.

The study design was developed using the PICO (participant, interventions, comparisons, outcomes) framework (31).

Only studies which investigated adherence interventions in India, in outpatient and community populations with at least one eligible non-communicable chronic disease, were considered. NCDs appropriate for inclusion were identified previously in the scoping search and based on those surveyed in the recent SAGE-2 report in India: stroke, angina pectoris, diabetes mellitus, asthma, depression, hypertension, and chronic lung disease (32). We excluded communicable diseases, such as tuberculosis and HIV, as interventions to improve adherence to medications for these diseases are better characterized (24, 33–37). We also excluded arthritis, as regular medication use is not the mainstay of treatment (38, 39).

Only studies which tested an intervention to improve adherence in patients taking regular medications, either as a primary or secondary outcome, were considered. Examples of eligible interventions were patient education, streamlined medication regimens, and electronic reminders. Studies only investigating interventions targeted at improving medication adherence within a hospital setting were excluded. Adherence outcomes included subjective data based on patient self-reporting and objective measures, such as pill counts. Studies which measured objective clinical parameters and disease outcomes as a proxy for adherence were also included if adherence to treatment was explicitly stated as a study aim.

A comprehensive search strategy, based on the above criteria and the scoping review, was devised. The search strategy was conducted on Ovid Medline, Web of Science, Scopus, and Google Scholar. No restrictions were placed on language, date of publication or geographical region. The full search criteria can be found in Supplementary material. Medical subject headings (MeSH), free-text terms, and Boolean logic were used where available. Searches were run on 20/7/2022. Given the comprehensive search strategy, references of included articles were not checked.

The studies retrieved by the search strategy, along with study information and abstract text, were imported into Zotero and then Mendeley for de-duplication (40). A 1,264 unique articles were split amongst reviewers, with each title and abstract screened independently by two reviewers using Rayyan AI (41). Conflicts between the two reviewers were resolved by a third independent reviewer. To ensure the inclusion and exclusion criteria were applied consistently amongst all reviewers, all reviewers screened 50 studies first and any disagreements were discussed by all reviewers until a consensus was reached. Full texts of included studies were screened independently by two reviewers; a third independent reviewer resolved any disagreements between the two original reviewers. For reports where full text was unable to be retrieved, authors were contacted to request their manuscripts—if they failed to respond within 2 months, studies were excluded.

Only randomized control trials were included in qualitative synthesis to ensure conclusions were informed by the best evidence available. Inclusion and exclusion criteria are shown in Table 1.

Data extraction was conducted via Google Sheets using a pre-specified template, using a list of features identified during the scoping review. One reviewer performed data extraction for each study, and this was subsequently checked by a second reviewer. The different domains of data extracted are reported in Supplementary material.

The Cochrane Risk-of-Bias (RoB) 2 tool for randomized control trials and for cluster randomized control trials were used to assess methodological quality (42). RoB assessment was conducted independently by two reviewers for each article, and any disagreements across domains were resolved by a third reviewer. Results were synthesized and formatted using R Studio (43). When multiple papers reported different adherence-related outcomes from the same original study dataset, study weights were adjusted accordingly to prevent double-counting.

We identified 22 RCTs which met the inclusion criteria (44–65). Of these 22 studies, 12 investigated the effects of patient education (mean length of follow-up = 6 months); two studies investigated the use of electronic reminder-based systems (mean length of follow-up = 9 months); two studies investigated psychological-based approaches (mean length of follow-up = 3.5 months), single studies investigated medication regimen changes (length of follow-up = 12 months) and improved clinical practitioner competency (length of follow-up = 5 months), and four RCTs used multiple interventions (mean length of follow-up = 12 months). Studies all evaluated adult patients (with the exception of Grover et al. which focused on pediatric patients) (44) with a mean age of patients between 50–65 (with Raj et al. focussing on those over 60 exclusively) (60) of both sexes with Pradeep et al. focussing on only female patients (49). Studies collected their data between 2006 and 2019 where timeframes were given (Ponnusankar et al. was published in 2004) (45). An overview of all included studies can be found in Table 2 below. Many studies had multiple outcome measures; those most relevant to adherence and corresponding clinical outcomes are reported in Table 2.

Adherence can be measured directly, through objective parameters such as pill counts and subjective parameters such as self-rated adherence, or indirectly through reporting on changes in objective clinical parameters which should improve with adherence promoting interventions.

Fourteen out of the 22 RCTs used both direct measures of adherence e.g., questionnaires, self-reporting or pill counting (pills consumed/pills prescribed), and indirect measures through changes in clinically relevant parameters e.g., systolic blood pressure, HbA1c, or health-related quality of life as end-outcomes (44, 46, 47, 49, 50, 52–54, 57, 59, 60, 63, 64, 66). Four studies only used indirect measures through changes in clinically relevant parameters (51, 58, 62, 65) and three studies only used direct measures of subjective adherence (48, 55, 56). One study used both objective and subjective direct measures of adherence (45).

Of the 18 studies that used direct measures of adherence as end-outcomes, three used objective measures of adherence e.g., pill counting (all three studies) (45, 60, 66) and 15 used subjective measures of adherence, such as self-reporting (five studies) (45, 47, 50, 55, 63)or questionnaires (ten studies) (44, 48, 51–54, 56, 57, 59, 64). Questionnaires measuring adherence included the Beliefs about Medicines Questionnaire (BMQ), Morisky Medication Adherence Questionnaire (MMAQ), Medication Adherence Rating Scale (MARS), Medication Adherence Questionnaire (MAQ), and questionnaires specifically designed for the purposes of the relevant study. One study's-outcome was the number of patients advised to adhere to medications rather than medication adherence per se (46).

Out of 22 included studies, five were low risk of bias, (50, 52, 64–66) six were high risk (44, 49, 51, 54, 56, 57) and 11 were assessed as some concerns (14, 45–48, 53, 55, 58, 59, 62, 63) (Figure 2). Notably, two studies were deemed “high risk” and eight assessed as “some concerns” for bias in selection of reported results category (Figure 2). One study was deemed “high risk” and six studies were deemed “some concerns” for bias arising in the randomization process. This was primarily due to lack of reporting of how randomization occurred. One study was assessed as “high risk” while eight were assessed as “some concerns” for bias arising from deviations from intended interventions, mainly due to lack of information surrounding if and how participants and observers were blinded. We also found that only 12 out of 22 studies had registered their study protocol (47, 50, 52, 53, 55–57, 61–64, 66). Overall Risk of Bias is shown in Figure 3. Information on individual study blinding is shown in the Supplementary material.

Seven studies examined educational programmes delivered by pharmacists to improve adherence and all showed significant improvements in adherence outcomes assessed (44, 45, 48, 54, 57, 59, 65). The interventions were generally short (10–30 min), personalized sessions which could be delivered repeatedly, and were often combined with provision of written materials and lifestyle advice. Studies found improvement in both objective and subjective metrics of adherence utilizing different regimens of pharmacist led sessions. All identified studies showed some benefits although some studies (44, 45) did not conduct complete statistical tests of significance (44, 45) whilst others only followed patients up over 2 months (45, 48). All studies except one (57) were based at single centers and all had relatively low sample sizes.

For example, Sriram et al. found that pharmacist teaching comprising education, medication counseling, and lifestyle advice during baseline and follow-up visits led to an improvement in HbA1c values (p < 0.01) and diabetes-dependent quality of life (p < 0.01) after 8 months compared to usual care across 120 patients with type 2 diabetes (65). Similarly, Simon et al. also found that a verbal counseling session and leaflets from clinical pharmacists focussing on medication adherence and lifestyle advice led to improved HbA1c (p < 0.001) and medication adherence (p < 0.001) scores compared to conventional therapy with basic counseling across 97 patients with type 2 diabetes after 6 months (57).

These benefits were not limited to type 2 diabetes, with Sundararajan et al. utilizing a single 30-min pharmacist counseling session at discharge, addressing subjects such as adherence and lifestyle advice, in patients post-myocardial infarction (54). They found improved medication adherence (p = 0.0001) and cardiovascular clinical parameters such as blood pressure (p < 0.01) after 6 months compared to usual care across 154 patients. Further, Sathvik et al. found improved medication adherence scores across belief and recall domains (p < 0.05) in hypertensive patients given 4 pharmacist education sessions compared to control patients given leaflets alone in a cohort of 150 over 2 months (48). Ponnusankar et al. was the only study to measure adherence using an objective pill-count method and found improved adherence after 2 months in a group of 90 patients with various chronic diseases following pharmacist a counseling session compared to usual care (45). They found an increase in pill-count scores and self-assessed compliance although they did not carry out formal statistical analyses.

Renuga et al. demonstrated potential added value of repeated interventions, showing that verbal counseling and education leaflets provided at baseline and during 3 follow-ups led to further improvements in adherence (p < 0.001) and fasting blood sugar level (p < 0.01) for patients with diabetes after 3 months, compared to the improvements seen with counseling at baseline and follow up alone, across 400 patients with type 2 diabetes (59).

Uniquely, Grover et al. looked at a pediatric asthma cohort and found that a structured programme delivered to both parents and children focussing on themes such as asthma knowledge and medication use improved medication adherence to asthmatic medications and asthma control (p < 0.01) over 6 months compared to control given an information pack and usual care (44). However, their sample size was only 40 parent-child pairs and no statistical analysis on adherence scores were conducted.

Four studies examined education-based interventions delivered by community healthcare workers (CHWs), non-physicians who are able to perform certain health-related interventions, especially in more rural areas of India (49, 63, 64, 66). These individuals usually received training on how to deliver educational interventions and then visited patients, providing individualized advice, counseling, and encouragement to take medications. Three studies found improvement in adherence parameters, however in two, objective clinical parameters failed to reflect improvements in subjective adherence metrics (49, 66). One study found improved clinical parameters but failed to show improvement in adherence metrics (63). Three of these four studies were large multicentre studies focussing on cardiovascular disease.

Xavier et al. showed the benefits of CHWs in improving both adherence measures and corresponding clinical parameters; in a large multicentre trial they found CHWs were effective at increasing adherence to secondary prevention drugs (p = 0.006) for 806 patients with acute coronary syndrome and improved clinical cardiovascular parameters such as systolic blood pressure (p = 0.002) after 1 year through providing repeated counseling on lifestyle measures and education (64).

Gamage et al. conducted a large cluster trial with 1,734 participants with hypertension and found trained CHWs significantly improved hypertension control compared to usual care (p = 0.001) in 3 socioeconomically distinct regions through their delivery of 6 fortnightly sessions over 3 months to educate patients on their disease, medication, and lifestyle choices (63). However, they did not find significant changes in antihypertensive medication use, indicating this effect may have been primarily due to lifestyle changes.

Joshi et al. (a large cluster trial with 2,312 patients) and Pradeep et al. (randomized trial with 260 women in 6 rural villages) found CHWs were effective at improving adherence for cardiovascular risk factors (through 6 household visits over 12 months in which they ascertained and reinforced, p = 0.001) and depression (through twice monthly visits in which education and counseling on medication use was provided for 6 months, p < 0.01) compared to usual care respectively, but both did not find significant improvements in corresponding clinical parameters (49, 66). Joshi et al. also looked at adherence and BP 6 months after the intervention ended and found a rise in mean blood pressure and a drop in adherence (61).

Sadeghian et al. (314 patients across two hospitals) utilized a self-management education programme delivered for 2 hours a week over 2 weeks, by a multidisciplinary team to patients with T2DM (58). The intervention team consisted of an endocrinologist, diabetologist, public health expert, dietician, diabetes nurse educator and the investigator. They reported an objective improvement in HbA1c at the 6-months follow-up period (p < 0.001).

Sylaja et al. conducted a cluster RCT of 243 subjects investigating the effects of improved CHW competency on the secondary prevention of stroke over a period of six months (46). Their intervention consisted of four 45-min training sessions for CHWs focused on the management of acute stroke, drug adherence, care, and physiotherapy. At 6 months follow-up there was an increase in the number of patients advised on medication adherence by CHWs in the intervention group compared to CHWs in the control group (p < 0.001), but this was not associated with significant improvements in blood pressure or fasting blood sugar in the intervention group.

Four studies (based on 3 RCTs) focused on multimodal interventions by non-physicians, providing education and counseling in combination with regular follow-up interventions (52, 56, 60, 62). All showed improvements in medication adherence metrics although one did not measure clinical parameters (56) although two showed no improvement in clinical outcome measures (52, 60).

Abdulsalim et al. conducted a single-center RCT looking at the benefits pharmacist-led counseling for 260 patients with COPD over a period of 2 years, in combination with monthly telephone calls, to ensure adherence (56). Pharmacist sessions focused on medication adherence and education, as well as lifestyle information, and leaflets were also provided. They found significant improvements in subjectively rated medication adherence (p < 0.001) after 2 years. The study was based on the results of a study by Suhaj et al. who independently reported significant improvement in health-related quality of life (p < 0.001) in the same cohort (62).

Sheilini et al. conducted a single-center RCT, investigating the benefits of a nurse-led intervention for 160 hypertensive patients non-adherent to medications over a period of 6 months (52). The intervention involved personalized education, information leaflets in combination with weekly reminder boxes and a single telephone follow-up. They reported improved adherence (p < 0.001) but no significant reduction in blood pressure (p = 0.16) for the intervention group compared to controls, which may have been due to unmodified lifestyle factors.

Raj et al. looked at tailored advice delivered by a study investigator (such as education if there was poor knowledge) combined with diary and mobile phone reminders in 50 older patients with a range of NCDs (60). They found a significant increase in reported pill counts after 3 months (p = 0.007) and 6 months (p = 0.003) although no significant changes in corresponding clinical parameters. However, they found a decrease in reported pill counts at 6 months relative to 3 months (p = 0.016).

Two studies investigated the effects of reminder-based systems on medication adherence (47, 51).

Kleinman et al. assessed the impact of a Gather Health, a mobile-Health diabetes management platform on medication adherence and frequency of blood glucose self-testing at three centers across 91 patients with type 2 diabetes (47). After 6 months follow-up, they found that the intervention group had significantly improved self-reported medication (p = 0.03), increased frequency of blood glucose self-testing (p = 0.01) and improvement in HbA1c compared to control (p = 0.02). Similarly, Shetty et al. conducted a single-center pilot study to investigate the effectiveness of mobile short message services on adherence in 215 diabetic patients (51). Messages were sent once every 3 days, and both content and frequency varied based on subjects” preferences. At 1 year follow-up they reported no significant difference in mean HbA1c values between both groups (p value not stated) but found a significant increase in the percentage of subjects with HbA1c < 8% in the intervention group compared to controls (p < 0.007). They stated that drug prescriptions were followed satisfactorily by both groups although which specific questionnaire was used was not stated.

Thom et al. conducted a large multi-center trial across 1,000 patients investigating the effects of fixed-dose-combination based strategies (polypills) on self-reported adherence and changes in blood pressure in those with cardiovascular disease or high risk of cardiovascular disease (the UMPIRE trial) (50). At follow-up (median = 15 months) those in the intervention group taking a polypill with their CVD medications were found to have improved self-reported medication adherence compared to usual care (p < 0.001). Improvements in adherence were associated with improvements in clinical parameters such as reductions in systolic blood pressure (p < 0.001) and LDL-C (p < 0.001).

Valsaraj et al. found that weekly 50-min cognitive behavioral therapy (CBT) sessions for 80 patients with chronic kidney disease undergoing dialysis, were associated with increases in dialysis adherence (p = 0.001) and drug adherence (p < 0.001) compared to non-directive counseling (53). These changes in adherence were associated with concomitant changes in objective parameters such as systolic blood pressure (p = 0.001), diastolic blood pressure (p = 0.001), Hb (p =0.001), and inter-dialysis weight gain (p = 0.001).

Pillai et al. reported on a collaborative stepped care intervention to improve adherence to anti-depressants across 2,796 patients with depression (55). The intervention involved psychoeducation, antidepressants prescribing, and interpersonal psychotherapy delivered according to illness severity and patient response. Compared to normal care (which included increased access antidepressant prescribing), they found improved self-reported adherence over a course of a month (OR = 6.10, 95% CI 3.67–10.14) and a significantly higher proportion completed at least 90 days of antidepressants, compared to usual care (although exact p values are not stated).

Table 3 shows a summary of the evidence for adherence promoting measures broken down by disease. Of the studies retrieved, three studies examined respiratory disease, six focussed on type 2 diabetes, eight on cardiovascular disease, two on depression, one on renal disease and 2 looked across NCDs (not included in Table 3).

Several studies found significant improvements in subjective adherence metrics without corresponding changes in objective clinical parameters (46, 49, 52, 60, 66). This may have reflected inadequate pharmacological therapy, emphasizing the importance of optimizing medical management, such as dose titration or additions of second line medications, in combination with adherence-promoting interventions (66, 84–86). It may also reflect trials that did not assess concomitant lifestyle changes alongside improving medication adherence; the potential for CHWs and pharmacists to counsel on weight loss, exercise and smoking cessation represent effective additional methods of improving health outcomes (52). Finally, small study sizes, limited time frames and controls which encouraged patients to seek additional care may also be responsible for a proportion of this effect (49, 60). We recommend future interventions aiming to improve medication utilize regular reviews to optimize a patients' pharmacological management and involve simultaneous counseling and assessment of lifestyle measures.

This review's conclusions are limited by mixed methodological quality in included studies. Moderate and high risk of bias across studies was primarily due to bias in selection of reporting results, lack of information on how randomization occurred, and if and how participants and observers were blinded. To mitigate reporting bias, we recommended that future RCTs pre-register their study protocol online. Additionally, more information is required on randomization and blinding processes used to ensure high methodological quality.

Additional limitations relate to the nature of the included studies. Firstly, there is volunteer bias to which all RCTs are subject, meaning it is unclear how generalizable these findings are to those who are more vulnerable or with lower health literacy. Secondly, most included studies primarily used subjective measures of adherence such as validated questionnaires or self-reporting, while only three employed objective measures of adherence. Subjective measures are liable to recall bias and social desirability bias, generally resulting in over-reporting and may limit the generalizability of our results, although a range of these questionnaires have been found to be valid across NCDs (87–90). Thirdly, of the available studies, there were few large multicentre RCTs and in general follow-up periods were short (median follow-up was 6 months). As medication adherence rates tend to decline with time from intervention possibly due to participant fatigue and waning novelty, follow-up mechanisms and evaluating for sustained effects are crucial to determining the efficacy of adherence promoting measures (24, 60, 66). Fourthly, given the relatively small number of included studies (22) across a large number of states and regions within India it was not possible to give a breakdown by state, where different healthcare systems and socioeconomic factors may influence adherence barriers and interventions. Finally, the COVID-19 pandemic has driven a shift that has rapidly changed the field of medication adherence, so previous studies (15/22 were published before 2020) may lack current day validity, especially for interventions dependent on technology such as mHealth (91).

Future large RCTs with long follow-up periods to test for sustained improvements in adherence, utilizing methods that attenuate volunteer bias (for example methods that increase recruitment rates), are required to further evaluate adherence-improving measures (92, 93). We recommend future studies evaluate adherence-improving measures using a variety of corresponding adherence and clinical metrics, utilizing objective measures of adherence where possible alongside simpler subjective measures.