Antidepressant Use and Cognitive Deficits in Older Men: Addressing Confounding by Indications with Different Methods

Abstracts from this study were presented at the American Association for Geriatric Psychiatry 2009 Annual Meeting, Honolulu, Hawaii, March 5–8, 2009 and the 12th International Conference on Alzheimer’s Disease, Vienna, Austria, July, 11–16 2009, with partial support form a travel fellowship provided by the Alzheimer’s Association.
https://doi.org/10.1016/j.annepidem.2011.10.004Get rights and content

Purpose

Antidepressant use has been associated with cognitive impairment in older persons. We sought to examine whether this association might reflect an indication bias.

Methods

A total of 544 community-dwelling hypertensive men aged ≥65 years completed the Hopkins Verbal Learning Test at baseline and 1 year. Antidepressant medications were ascertained by the use of medical records. Potential confounding by indications was examined by adjusting for depression-related diagnoses and severity of depression symptoms using multiple linear regression, a propensity score, and a structural equation model (SEM).

Results

Before adjusting for the indications, a one unit cumulative exposure to antidepressants was associated with −1.00 (95% confidence interval [CI], −1.94, −0.06) point lower HVLT score. After adjusting for the indications using multiple linear regression or a propensity score, the association diminished to −0.48 (95% CI, −0.62, 1.58) and −0.58 (95% CI, −0.60, 1.58), respectively. The most clinical interpretable empirical SEM with adequate fit involves both direct and indirect paths of the two indications. Depression-related diagnoses and depression symptoms significantly predict antidepressant use (p < .05). Their total standardized path coefficients on Hopkins Verbal Learning Test score were twice (0.073) or as large (0.034) as the antidepressant use (0.035).

Conclusion

The apparent association between antidepressant use and memory deficit in older persons may be confounded by indications. SEM offers a heuristic empirical method for examining confounding by indications but not quantitatively superior bias reduction compared with conventional methods.

Introduction

The authors of observational studies of adverse drug effects face a challenge in ascertaining the causation of an observed association, i.e., confounding by indications or reasons for prescribing. When an adverse health outcome is observed, the drugs prescribed to treat a disease may be claimed as the cause, whereas in fact the underlying disease that triggered the drug treatment is responsible 1, 2, 3. The source of this bias can be traced to a study design defect, i.e., the lack of randomization in treatment assignment. As a result, an observed “treatment effect” may reflect the differences in key population characteristics that affect both the probability of receiving the treatment and risk of the outcome, rather than the treatment under study. Because of the inability to pre-balance these characteristics at the design stage, observational studies must address such a potential bias analytically 1, 2.

The most common analytical method to control confounding by indications is to use a multivariable method, such as multiple linear regression (MLR), in which the identified “indications” or their surrogates are included as covariates. The adjusted effect estimate for drug use is then conditional on the observed values of these covariates or as a weighted average of them 1, 2 and hence, free of indication bias. Although such a direct modeling adjustment is simple and intuitive, treating the indication as regular confounders has several limitations. First and foremost, the temporal precedence of the indications to the drug use can not be realized in a conventional regression context, where both are treated as equally important independent variables. In addition, because of a lack of specific clinical assessments, observational studies often have to rely on multiple nonspecific measures as surrogates for indications for drug prescribing, which makes regression adjustment scientifically dubious and statistically inefficient 2, 3, 4.

To move towards causality, a two-step propensity score (PS) method was developed 2, 4, 5. First, a PS, or conditional probability, of receiving a specific drug given a set of predictors is estimated. Then in the hypothesis testing stage, the estimated PS is used as a covariate, or a stratification or matching variable, to control for confounding in the multivariable model of the relationship between drug exposure and study outcome. Conceptually the PS is expected to increase the validity (by pre-balancing observed confounders between exposure groups) and efficiency (by reducing number of parameters) of regression models. However, studies have not always found it superior to traditional multivariable adjustments (5).

Path analysis is a type of structural equation modeling (SEM) that tests statistical models among observed variables by the use of covariance structures. Although sharing many properties with traditional MLR, path analysis has several unique features. First, it allows for explicit specification of the directionality and hierarchy of several hypothesized “causal” relationships 6, 7, 8. Second, it recognizes the imperfect nature of measurements by directly modeling, rather than ignoring, measurement errors, and by providing multiple model fit statistics, including model χ2, root mean squared error of approximation (RMSEA) and comparative fit index (CFI). Finally, it typically uses a graphical path diagram as a starting point to delineate theoretically complex (causal or noncausal) relationships. Estimates for the specified relationships and model fit are then derived by solving a set of simultaneous equations. Unlike the MLR and PS modeling, the use of SEM in the medical field is relatively limited 7, 9.

In this study, we illustrate how conventional MLR, PS, and SEM can be used to address potential confounding by indication surrounding the relationship between antidepressant use and cognitive impairment by using data from a cohort of community-dwelling older persons. Antidepressant medications, especially the tricyclic agents, have been associated with cognitive impairment in the elderly 10, 11. Although recent randomized trials demonstrated potential cognitive benefits of antidepressant medications in treating major depressive disorders 12, 13, little is known about their long-term cognitive outcome in older persons living in the community, who often suffer from mild-to-moderate depressive symptoms, but multiple other chronic conditions 14, 15. Our hypothesis was that the apparent relationship between antidepressant use and cognitive impairment observed in the elderly may reflect potential confounding by underlying depression or the severity of depressive symptoms and would therefore diminish after appropriate adjustment for such covariates.

Section snippets

Participants

The Connecticut Veterans Longitudinal Cohort consisted of 767 community-dwelling older persons aged ≥65 years. The cohort was originally assembled at a Veterans Administration (VA) primary care clinic between July 2000 and August 2001. Data on demographic profiles, health-behaviors, comorbidities, blood pressure readings, and cognitive and physical functioning were collected at baseline and at 1-year follow-up. The study protocol was approved by the Yale University School of Medicine

Baseline Characteristics and Antidepressant Use During Follow-Up

At baseline, the study cohort had mild-to-moderate hypertension severity, a high level of functional autonomy, and a lower level of depressive symptoms (Table 1). One hundred twenty (22.1%) participants had a depression-related diagnosis, and 85 (15.6%) were prescribed antidepressants. The most frequent antidepressants are sertraline (5.2%), amitriptyline (2.8%), nortriptyline (1.8%), trazodone (1.5%), and fluoxetine (1.3%). During the 1-year follow-up period, 107 (19.7%) participants were

Discussion

In this cohort of community-dwelling older men with hypertension, we demonstrated that the apparent association between antidepressant use and memory deficit diminished substantially and became nonsignificant after adjusting for potential indications for antidepressant prescribing, or a PS for it, but not by using the global comorbidity measures and other risk factors alone. In addition, the best fitting empirical SEM with greatest clinical interpretability seems the one under our a priori

References (36)

  • M. Salas et al.

    Confounding by indication: An example of variation in the use of epidemiologic terminology

    Am J Epidemiol

    (1999)
  • R.J. Glynn et al.

    Indications for propensity scores and review of their use in pharmacoepidemiology [review]

    Basic Clin Pharmacol Toxicol

    (2006)
  • R.B. Kline

    Principles and Practice of Structural Equation Modeling

    (2004)
  • R.C. MacCallum et al.

    Application of structural equation modeling in psychological research

    Ann Rev Psychol

    (2000)
  • D.P. MacKinnon et al.

    Mediation analysis

    Ann Rev Psychol

    (2007)
  • S. Cadell et al.

    Factors contributing to posttraumatic growth: a proposed structural equation model

    Am J Orthopsychiatry

    (2003)
  • T.E. Oxman

    Antidepressants and cognitive impairment in the elderly

    J Clin Psychiatry

    (1996)
  • M.A. Butters et al.

    Changes in cognitive functioning following treatment of late-life depression

    Am J Psychiatry

    (2000)
  • Cited by (7)

    • Antidepressants and risk of dementia in migraine patients: A population-based case-control study

      2017, Progress in Neuro-Psychopharmacology and Biological Psychiatry
      Citation Excerpt :

      The use of SSRIs, TCAs, and NGAs might also be markers for disease severity or resistance to treatment. This type of confounding by indication has been observed in a study of antidepressant exposure and cognitive impairments in older men (Han et al., 2012). Second, there may be fundamental differences in the activity of different antidepressants depending on the underlying neuropsychiatric condition.

    • A systematic review of the statistical methods in prospective cohort studies investigating the effect of medications on cognition in older people

      2016, Research in Social and Administrative Pharmacy
      Citation Excerpt :

      Mura et al23 extended the use of linear mixed-effects models to evaluate the influence of covariates on a latent cognitive trajectory, where the latent cognitive outcome was characterized by nonlinear models that incorporated covariates. Other advanced regression techniques included generalized linear mixed models,24 generalized estimating equations20 and structural equation models.25 The use of advanced techniques (predominantly linear mixed-effects models) increased over time.

    • The problem of controlling for imperfectly measured confounders on dissimilar populations: A database simulation study

      2014, Journal of Cardiothoracic and Vascular Anesthesia
      Citation Excerpt :

      We do not mean to imply that misclassification bias, sometimes also referred to as information bias, is the only type of bias that afflicts studies attempting to compare group assignment on outcome. Particularly germane to large observational studies, other sources of bias include selection bias,26 confounding by indication,27 and time-dependent confounding by treatment,28 each of which has played an important role in other work that has attempted to define the treatment effects of particular drugs on an outcome of interest. Moreover, as the sensitivity analyses above demonstrate, propensity-score matching is not uniquely susceptible to misclassification bias.

    View all citing articles on Scopus

    This study was supported in part by the Claude D. Pepper Older Americans Independence Center at Yale University School of Medicine (#P30AG21342) from National Institute on Aging. The funding source had no involvement in the study design, data collection, analysis and interpretation.

    The views expressed in this article are those of the authors and do not necessarily reflect the position or policy of the Department of Veterans Affairs.

    View full text