Depressive symptoms and non-adherence to treatable cardiovascular risk factors’ medications in the CONSTANCES cohort Free

Abstract

Introduction

Depression is not only one of the leading causes of disability worldwide, but it has been repeatedly associated with cardiovascular disease (CVD).^1,2 In addition, cardiovascular risk factors are highly prevalent in the presence of depression.^3–5 The association between depression and CVD could thus be partially explained by poor adherence to medications regarding treatable cardiovascular risk factors, namely hypertension, diabetes mellitus, and dyslipidaemia. The prevalence of non-adherence to medications in chronic diseases, including hypertension, diabetes, and dyslipidaemia,⁶ is estimated to be 50% in developed countries.⁷ Moreover, around 20% of newly prescribed medications are never filled, and around 50% of those that are filled are not taken correctly in terms of frequency, dosage, duration, and timing.⁸ As hypertension, diabetes, and hypercholesterolaemia contribute to 68% of all deaths worldwide,⁶ improving adherence to medications treating these conditions is a critical challenge for public health.^9,10 This challenge might be even more critical in individuals with depression.

Overall, depression has been associated with poor adherence to medications in various settings.¹¹ A recent meta-analysis showed that depressed patients are 1.76 times more likely to be non-adherent to chronic disease medications compared to their non-depressed counterparts. Specifically, the adjusted odds ratios (ORs) of being non-adherent among depressed patients are 1.73 [95% confidence interval (CI): 1.24–2.87] and 1.79 (95% CI: 1.28–2.51) compared to non-depressed patients in six studies on diabetes and eight studies on hypertension or dyslipidaemia, respectively.¹¹ However, adjustment variables differ from one study to the other and residual confounding could not be excluded. In addition, most of these studies were conducted in clinical samples and evidence from the general population is sparse. To our knowledge, eight studies assessed these associations using population-based samples, but only three were prospective studies using objective measures of treatment adherence^12–14 and none of these three studies assessed hypertension, diabetes, and hypercholesterolaemia jointly.

Taking advantage of a population-based cohort study linked to administrative claim database, we aimed to examine the prospective association between depressive symptoms and subsequent non-adherence to medications each targeting hypertension, diabetes type 2 and dyslipidaemia in a large-scale, diverse adult population living in France.

Methods

Study design

The CONSTANCES cohort is a large population-based study of 18–69 years old volunteers randomly selected from 22 health-screening centres (HSCs) in different districts across France since 2012 that reached 200 000 participants by early 2019.¹⁵ These volunteers were covered by the French national health insurance fund; hence, this study covered 85% of the adult population in France and excluded self-employers, farmers, and undocumented immigrants. Volunteers were asked at inclusion to complete self-administrated questionnaires by mail on lifestyle and socio-demographic characteristics as well as socio-professional and health statuses. Additionally, they underwent health examination such as laboratory tests and physical examination in the HSCs. If a new disease was discovered after the full assessment, including hypertension, diabetes, and dyslipidaemia, the participants were informed and offered a medical consultation to review their results with the HSC physicians.

Participants’ data were matched to the ‘Système National d’Information Interrégimes de l’Assurance Maladie’ (SNIIRAM) database through specific identifiers. This database included information on all reimbursed medications,¹⁶ which served to identify the use and adherence to pharmacological treatments targeting several diseases including cardiovascular risk factors such as hypertension, diabetes, and dyslipidaemia from 1 January 2009 to 31 December 2017.

The CONSTANCES cohort received authorization from regulatory bodies supervising medical research in France which are the ‘Commission de l’Informatique et des Libertés—CNIL’ (authorization no. 910486), the Institutional Review Board of the National Institute for Medical Research-INSERM, the ‘Conseil National de l’Ordre des Médecins—CNOM’ and ‘Conseil National de l’Information Statistique—CNIS’.¹⁵

Study populations

We first defined the presence of hypertension, diabetes type 2, and dyslipidaemia by either self-reports in the medical questionnaire administrated by the physician at inclusion, or abnormal blood pressure/blood tests on the medical examination offered at study inclusion, or both. Abnormal blood pressure was defined as a systolic blood pressure of ≥140 mmHg and/or diastolic blood pressure of ≥90 mmHg. Abnormal glycaemia, cholesterol, triglycerides levels were defined as fasting glucose ≥7 mmol/L, fasting LDL-cholesterol >4.08 mmol/L, and/or fasting triglycerides ≥1.7 mmol/L, respectively. Blood pressure and blood tests were collected from laboratory tests at inclusion during the visit to the HSCs based on standardized procedures.¹⁷ For instance, blood pressure was measured using oscillometric sphygmomanometer based after 5 min of rest. The blood tests were collected and pre-treated within a maximum of 30 min period, then transported to the central laboratory on the same day at 4–8 degrees.¹⁵

A total of 151 207 participants were enrolled in CONSTANCES between February 2012 and 1 March 2018. Among them, 128 700 had available SNIIRAM data and 45 052 were included before 1 January 2015, thus having 36 months of follow-up (Supplementary material online, Figure S1). Among these 45 052 individuals, we included individuals with the condition under study (i.e. either hypertension, diabetes, or dyslipidaemia), who had at least one drug prescriptions filled in the 6 months before inclusion and who completed the Center for Epidemiologic Studies Depression (CES-D) scale at inclusion. Since our primary outcome was medication non-adherence at follow-up, being on medication on the day of inclusion was not required as inclusion criteria. Information on date of drug prescriptions filled was collected from the SNIIRAM database. Pharmacological treatment of the conditions under study was identified according to the World Health Organization’s Anatomical Therapeutic Chemical (ATC) classification.¹⁸

Assessment of depressive symptoms

Depressive symptoms were assessed at baseline using the French version of the CES-D scale.¹⁹ The CES-D scale is composed of 20 items that assess depressive symptoms in the past week such as agitation, guilt, loss of appetite and interest, fatigue, suicidal thoughts and sadness (e.g. I felt sad, I felt lonely) on a scale from 0 to 3 with 0 being ‘rarely’ (<1 day) and 3 ‘most of the time’ (5–7 days). Internal consistency of CES-D scale is generally high (α = 0.90 in the CONSTANCES cohort). Volunteers with a score of ≥19 were considered to be clinically depressed according to the validated threshold for the French version (sensitivity and specificity >85% for the diagnosis of major depression).¹⁹ However, to incorporate volunteers with more severe depressive symptoms for minimizing type I error, sensitivity analyses were performed with the CES-D score using two thresholds with higher specificity: 21 (sensitivity 82%, specificity 91%) and 22 (sensitivity 79%, specificity 92%).¹⁹

Assessment of non-adherence to medications

Non-adherence to cardiovascular risk factors’ medications under study was assessed by tracking the reimbursement of ad hoc medications over 18 months after the date of inclusion and 18 months afterwards (19–36 months). To measure medication non-adherence, we divided the sum of trimesters with at least one drug dispensed by the total number of trimesters (i.e. 6 trimesters)²⁰ (Supplementary material online, Equation S1). Participants were considered non-adherent if they purchased the studied drugs less than 80% of time, a commonly used threshold in studies of medication non-adherence.²⁰ In addition, exploratory analyses were performed using medication non-adherence between the first 3–6 months after the date of inclusion (i.e. had at least one trimester without drug dispensed) to test whether non-adherence to medications would be more sensitive to depressive symptoms in this period.

Assessment of covariates

Age (continuous, years), sex, living with a partner (yes; no), country of birth (France; other), educational level (≤ high school diploma: ≤13 years of education; undergraduate degree: between 14 and 16 years of education; post-graduate degree: ≥17 years of education), and employment status (yes; no) were collected at inclusion from volunteers’ self-reports. Physical activity (0: not active to 6: very active) and prescribed and followed diet (yes; no) at inclusion were also collected from the same source for descriptive purpose. Personal history of CVD (yes: history of angina pectoris, myocardial infarction, stroke, or peripheral arterial diseases; no) was collected from the medical questionnaires administrated by the physician at inclusion. Body mass index (BMI) (<25; ≥25 and <30; ≥30 kg/m²) was calculated at inclusion.

Statistical analysis

All analyses were conducted using the SAS system software (version 9.4, SAS Institute, Cary, NC, USA). Independent t-test, χ² tests, and Fisher’s exact tests were computed for demographic and clinical characteristics by the presence of depressive symptoms. First, as our primary analyses, the associations between depressive symptoms at inclusion and non-adherence to medications treating hypertension, diabetes, or dyslipidaemia either in the first 18 months or between 19 and 36 months of follow-up were studied with unadjusted and fully-adjusted logistic regression models. Based on an anticipated rate of non-adherence of 25% in this sample of individuals presumably more health-conscious than the general population, an anticipated rate of 20% of participants with depressive symptoms (i.e. a CES-D score >19) and an anticipated OR of 1.75 for non-adherence in the presence of depressive symptoms,¹¹ we calculated that a minimum of 706 participants would be needed for a given condition to achieve a statistical power of 80% with an alpha risk set at 5%.²¹ This sample size was corresponding to the expected sample size of participants with diabetes, with larger samples being expected for participants with hypertension or dyslipidaemia. Fully-adjusted models were adjusted for all the covariates mentioned above, except for prescribed diet and physical activity. These two variables were excluded since they could be part of the non-pharmacological treatment of the condition, thus capturing a part of medical adherence. In other words, adjusting for these variables may result in an artificial underestimation of the association of depressive symptoms with medication non-adherence. Second, as exploratory analyses, these analyses were repeated between the first 3–6 months to explore whether non-adherence to medications would be more sensitive to depressive symptoms in this period.

Since socio-demographics (i.e. age, sex and educational level) may be effect modifiers of the association between depressive symptoms and non-adherence to medications, we tested for statistical interactions between depressive symptoms and these variables. Should these interactions be significant, we planned to further examine the association between depressive symptoms and non-adherence to medications in stratified analyses.

All analyses were computed using multiple imputations to handle missing values in the covariates.

Results

The characteristics of participants with hypertension (N = 4998), diabetes (N = 793), and dyslipidaemia (N = 3692) are presented in Table 1. The overlap between the three populations under study is displayed in Supplementary material online, Figure S2. The prevalence of depressive symptoms varied from 16.2% in individuals with dyslipidaemia to 21.8% in those with diabetes. The majority of study participants were men, living with a partner, French by birth, non-employed, physically active, had a high school diploma at best and a BMI between 25 and 30, did not follow a prescribed diet, did not have a personal history of CVD. Over the first 18 months of follow-up, 0.2% of individuals with hypertension, 0.4% with diabetes and 0.1% with dyslipidaemia died and thus were not included in the subsequent analyses as well as 0.9, 1.4, and 0.4% between 19 and 36 months.

Among individuals with hypertension, diabetes, and dyslipidaemia, respectively 13.1% vs. 11.5%, 10.5% vs. 5.8%, and 29.0% vs. 27.1% of those with vs. without depressive symptoms were non-adherent (i.e. had a proportion of trimesters with at least one drug dispensed of <80%) over the first 18 months of follow-up, and 15.9% vs. 13.6%, 11.1% vs. 7.4%, and 34.8% vs. 36.6% between 19 and 36 months.

Among participants with diabetes, hypertension, or dyslipidaemia, depressive symptoms were neither associated with non-adherence to medications over the first 18 months of follow-up, nor afterwards (Table 2). Similar findings were obtained in the fully adjusted models while using CES-D thresholds of 21 and 22 (Supplementary material online, Tables S1 and S2). None of the interactions between depressive symptoms and socio-demographic variables were significant (all P ≥ 0.06); hence, we did not further stratify our analyses according to these variables. However, in the exploratory analyses, depressive symptoms became associated with non-adherence to anti-diabetic medications between the first 3–6 months of follow-up with ORs ranging from 2.40 to 2.80 whatever the CES-D score threshold (Supplementary material online, Table S3).

Discussion

This study aimed to examine the prospective association between depressive symptoms, and non-adherence to medications for treatable major cardiovascular risk factors, namely hypertension, diabetes type 2, and dyslipidaemia, in a population-based setting. Overall, the results suggest that depressive symptoms were neither associated with non-adherence to medications for hypertension, diabetes, and dyslipidaemia over the first 18 months, nor afterwards. However, depression might be associated with non-adherence to antidiabetic medications for diabetes between the first 3–6 months.

There were certain strengths in this study. First, to our knowledge, this is the first study that examines the association between depressive symptoms and non-adherence to medications targeting the three main treatable cardiovascular risk factors in a single prospective, population-based cohort using administrative claim databases rather self-reported adherence. Second, the systematic linkage between a national population-based cohort and a national healthcare administrative database offered the opportunity to use actual prescriptions filled as dependent variable with no loss to follow-up while considering several potential confounders or effect modifiers. Overall, except for educational level, the significant associations of these potential confounders with non-adherence to medications were somehow consistent with those found in the literature.^22–25 Third, the clinical and laboratory data were collected according to standardized procedures by specialized teams at HSCs.¹⁷ Fourth, because the results of these procedures were disclosed to the participants, non-adherence is unlikely to be explained by a lack of awareness of the disease.

However, this study had also limitations, mainly resulting from the characteristics of the SNIIRAM database. First, this database does not provide detailed information on the conditions warranting the prescription of medications. For instance, some anti-diabetic medications may have been prescribed to lose weight.²⁶ Second, the SNIIRAM database does not contain information about actual medication use. However, it is unlikely that patients with regularly filled prescriptions did not take their medication at all. Third, detailed information about non-pharmacological treatments is not available in this database. For instance, participants who did not take pharmacological treatment were considered to be non-adherent, although some of them were possibly prescribed non-pharmacological treatments (e.g. change in diet, increased physical activity) only.^27,28 Fourth, although the CONSTANCES cohort is a large community-based population in France, with a broad socio-demographic diversity, the majority had a favourable social context. In addition, the prevalence of the conditions was lower than expected from the general population and this could be explained by the fact that CONSTANCES participants are more interested in their health. However, the low prevalence of diabetes is a very common phenomenon in any cohort²⁹ where individuals with diabetes are typically less prone to participate than to those with hypertension and dyslipidaemia. Furthermore, we showed that socio-demographics were not significant effect modifiers, thus limiting the risk of our results to be affected by selection biases. Fifth, the CES-D scale measures self-reported depressive symptoms and not major depression. For instance, functional impairment because of depressive symptoms is a diagnostic criterion of major depression that is not assessed by the CES-D. However, the association between depression and increased risk of CVD is not specific to major depression but also observed for subthreshold depressive symptoms. Our hypothesis was therefore that current depressive symptoms could impede medication adherence. However, since these symptoms were assessed only once at baseline, they may reflect a transient state that was not present later during follow-up. Sixth, we did not take into account antidepressant medications because, as stated above, our hypothesis was about current depressive symptoms and antidepressants per se were not supposed to have a direct effect on medication adherence. In addition, data from the French national health insurance system suggest that antidepressant drugs are frequently prescribed for reasons other than depression (i.e. migraine).³⁰ Finally, although the broader hypothesis we aimed to test was whether medications adherence to treatable cardiovascular risk factors could be a potential mediator of the association between depressive symptoms and CVD, the present study only tested a part of this hypothesis.

The lack of association between depressive symptoms and non-adherence to medications for hypertension or dyslipidaemia in the participants concerned was not in line with our hypothesis. This lack of association was indeed consistent with the results of some previous studies^31–33 but not with others^34–38 for hypertension. It was also not in line with the results of one study for dyslipidaemia³⁹; however, the authors did not assess depressive symptoms by themselves: they combined it with anxiety. Overall, these studies were either retrospective^32,40 or prospective with a follow-up period of 2 years at best. Furthermore, only three studies^12–14 used prospective population-based cohorts to assess the medication adherence based on objective measures. These three studies showed a positive association between depressive symptoms and non-adherence to medications over time. However, the first study measured non-persistence rather than non-adherence.¹³ The second study assessed non-adherence to medications targeting various cardio-metabolic conditions among adults with diabetes¹² and the third one used either depressive symptoms or anxiety as independent variable among old adults with hypertension.¹⁴

In contrast, the lack of a significant association between depressive symptoms and non-adherence to medications in individuals with diabetes in our primary analyses is more difficult to reconcile with the literature.^41–47 Only two studies^48,49 showed no association between these two variables and mainly when medication adherence was assessed using electronic caps.⁴⁹ However, most studies linking depression with non-adherence to anti-diabetic medication were either retrospective studies^41–43 or prospective with only one timepoint at follow-up. Indeed, depressive symptoms were associated with an almost two-time higher odd of being non-adherent to anti-diabetic medications in the first 18 months, whereas the OR was close to one for antihypertensive or lipid lowering medications. However, compared to hypertension and dyslipidaemia, our study might have lacked statistical power for diabetes because of the lower than expected proportion of those being non-adherent to anti-diabetic medications (i.e. 7 instead of 25%). This proportion would have required a 50% larger sample to achieve a statistical power of 80%.²¹ Compared to hypertension and dyslipidaemia, the treatment for controlling diabetes is by far more demanding for patients, as they have not only to implement both pharmacological and non-pharmacological treatments but also to monitor the condition several times a day, as well as being aware of possible acute complications of both the disease (e.g. ketoacidosis) and treatment (e.g. hypoglycaemia) and to be prepared to manage them in first line; hence depressive symptoms have more room to eventually interfere with medication adherence. Finally, the effect of a potential association between depressive symptoms at inclusion and non-adherence to anti-diabetic medication may have reduced after 18 months of follow-up because depressive symptoms are a potentially transient state rather than a permanent trait. In support to this hypothesis, exploratory analyses showed that depressive symptoms were associated with non-adherence to anti-diabetic medications between the first 3–6 months of follow-up only. Hence, further studies may consider depressive symptoms as time-dependent variables to examine this hypothesis. Specifically, repeated measures of depressive symptoms would allow testing the hypothesis of a detrimental effect of depressive symptoms on adherence to anti-diabetic medications—or other demanding treatment—in the short run only with a within-individual approach. In addition, studies that would rely on data gathered in the general population rather in volunteers only are needed to provide more definitive conclusions regarding the lack of association between depressive symptoms and adherence to medications for hypertension or dyslipidaemia.

In conclusion, in a population-based cohort, depressive symptoms were not associated with non-adherence to medications for hypertension, diabetes and dyslipidaemia in primary analyses. Although we may have lacked power for anti-diabetic medications, our results suggest that non-adherence to medications targeting treatable cardiovascular risk factors is unlikely to explain much of the association between depressive symptoms and CVD incidence. Although this association has been established for years, ruling out the hypothesis of a mediation by medication non-adherence is critical for depressive symptoms to be considered as CVD risk factor on its own.² Our finding suggests that other pathways (i.e. pathophysiological pathways) should be further explored to understand this association and identify relevant targets for preventive strategies. For example, depressive symptoms could lead to autonomic dysregulation triggering increased risk of CVD as the cortisol levels are high in depressed individuals and they are associated with CVD outcomes.⁵⁰ However, exploratory results suggest that depressive symptoms may be associated with anti-diabetic medications non-adherence in the short run. Although this finding warrant replication, it suggests that searching and treating depression may be especially important in individuals with diabetes as it may contribute to foster medications adherence. Future studies based on repeated measures of depressive symptoms and adherence in larger samples may further test this hypothesis.

Supplementary material

Supplementary material is available at European Heart Journal – Cardiovascular Pharmacotherapy online.

Acknowledgements

The authors thank CNAM and its health-screening centres for collecting most of the data. They also thank the National Old-Age Insurance Fund for its contribution to the cohort establishment, and ClinSearch, Asqualab, and Eurocell for data quality control.

Funding

N.H. was supported by a grant from Association Philippe Jabre. C.L. was supported by a grant from IReSP (LEMOGNE AAP16_PREV-13). The National Health Insurance Fund (Caisse nationale d’assurance maladie des travailleurs salaries, CNAM), MSD, Lundbeck, AstraZeneca, and ANR (ANR-11-INBS-0002) supported the CONSTANCES cohort. None of these funding sources had any role in the design of the study, collection and analysis of data or decision to publish.

Data availabillity

Personal health data underlying the findings of our study are not publicly available due to legal reasons related to data privacy protection. CONSTANCES has a data sharing policy but before data transfer a legal authorization has to be obtained from the CNIL (Commission nationale de l’informatique et des libertés), the French data privacy authority. The CONSTANCES email address is [email protected].

Conflict of interest: F.L. declares personal fees and non-financial support from Lundbeck, and non-financial support from Janssen-Cilag and Otsuka Pharmaceutical, outside the submitted work. C.L. declares personal fees from Boehringer Ingelheim, Janssen-Cilag, Lundbeck and Otsuka Pharmaceutical, outside the submitted work. The remaining authors declare no conflicts of interest.

References