Anticholinergic Medication Burden–Associated Cognitive Impairment in Schizophrenia

As described by Swerdlow et al. (9), participants underwent detailed diagnostic and symptom assessments, including the Scale for the Assessment of Positive Symptoms (SAPS) (10), the Scale for the Assessment of Negative Symptoms (SANS) (11), and the Global Assessment of Functioning (GAF) (12). Chlorpromazine equivalents for antipsychotic dosages were determined as previously described (13). The demographic and clinical characteristics of participants included for analyses and the characteristics of medication classes prescribed to patients are summarized in Table 1.

Anticholinergic burden from medication regimens was calculated for each participant, using a modified version of the Anticholinergic Cognitive Burden (ACB) scale. The ACB scale is a validated expert rating scale of the anticholinergic properties of medications (4, 14–16) with established clinical utility for predicting dementia risk in large-scale longitudinal clinical outcome studies of healthy older subjects. The ACB scale assigns a dose-independent rating for each medication based on its anticholinergic properties: 1 for low/minimal activity, 2 for moderate activity, and 3 for strong/definite anticholinergic activity. ACB values assigned to various medications and their frequencies in our sample are reported in Table S1 in the online supplement. For medications for which published values were not available, we assigned scores on the basis of known structural or pharmacological similarities to medications with existing ACB scores or consensus values derived from similar scales. In this context, fluphenazine (N=49) was given a score of 3 because of ratings of high anticholinergic burden on the Anticholinergic Risk Scale and the Anticholinergic Load Scale (17, 18) and its similarity to perphenazine. Prochlorperazine (N=2) was given a score of 2 because of moderate anticholinergic burden as rated by the Anticholinergic Risk Scale and the Anticholinergic Load Scale. Thiothixene (N=4) was given a score of 2 because of moderate anticholinergic burden as reported by Duran (19). Ziprasidone (N=91) and lurasidone (N=14) were both given a score of 1 because of ratings of either low/minimal or no anticholinergic burden (17). Although not described by the original ACB scale as having significant anticholinergic properties, citalopram (N=47), escitalopram (N=33), fluoxetine (N=58), mirtazapine (N=21), and sertraline (N=72) were given a score of 1 because several other scales rated them as having minimal anticholinergic burden (18, 20, 21). Duloxetine (N=18) was given a score of 1 because of its class similarity to venlafaxine. Nefazodone (N=1) was given a score of 1 because of its structural similarity to trazodone. Because diazepam was assigned an ACB score of 1 in previous reports, alprazolam (N=5), clonazepam (N=59), clorazepate (N=1), flurazepam (N=2), lorazepam (N=47), oxazepam (N=1), and temazepam (N=10) were also given a score of 1, as was zolpidem (N=45). Because carbamazepine was previously rated with an ACB score of 2, oxcarbazepine (N=10) was also rated as 2. Valproic acid (N=122) was given a score of 1, based on low/minimal anticholinergic burden as described by the Anticholinergic Drug Scale (21). Total ACB scores were generated by summing the individual ACB values from all medications for each participant, consistent with established methods (4, 8, 14, 15, 22, 23). Participants were grouped into five categories for subsequent analyses: no anticholinergic burden (ACB score=0), low anticholinergic burden (ACB score=1 or 2), moderate anticholinergic burden (ACB score=3 or 4), high anticholinergic burden (ACB score=5 or 6), or very high anticholinergic burden (ACB score >6).

The Penn Computerized Neurocognitive Battery (PCNB) was used as the primary outcome measure for cognitive functioning (24). The PCNB includes accuracy and speed measures in eight domains: abstraction and mental flexibility, attention, working memory, face memory, verbal memory, spatial memory, spatial ability, and emotion processing, reported as age- and gender-corrected z-scores. Efficiency scores for these eight domains were obtained as the average of accuracy and speed scores, as has been previously described (24). A PCNB global cognition score was derived by averaging individual efficiency scores (24). The PCNB included additional tests of motor praxis and overall motor speed that do not contain accuracy components; these tests were analyzed separately. Aside from the PCNB, a number of other cognitive assessments were administered to COGS-2 participants, including the Continuous Performance Test–Identical Pairs Version (CPT-IP; 3- and 4-digit) (25), the Degraded-Stimulus Continuous Performance Test (DS-CPT) (26), the letter-number span (forward and reorder) (27), and the California Verbal Learning Test–II (list A trials 1–5, list A total, list B total, short-delay free and cued recall, and long-delay free and cued recall) (28).

The impact of cumulative anticholinergic medication burden (none, low, moderate, high, or very high) on individual tests of cognition described above was examined by one-way Bonferroni-corrected analysis of variance with the initial significance threshold set to 0.05. Follow-up exploratory covariate regression analyses were conducted to determine whether relationships between ACB score (as a continuous variable, range 0–20) and cognition persisted after controlling for other important variables, including antipsychotic dosage, number of antipsychotics, positive symptom severity, negative symptom severity, duration of illness, number of hospitalizations, and cigarettes smoked per day. Lastly, we used exploratory structural equation modeling (ESEM) to determine whether relationships between ACB score (continuous variable) and cognition were specific to individual cognitive domains in the PCNB. This data reduction approach allowed us to reduce associations between ACB scores and each of the cognitive measures into associations between ACB scores and a smaller number of primary, core cognitive domains (latent cognitive domain factors). The parallel analysis function in the R psych package was used to determine the number of primary cognitive domains measured by the PCNB. Next, the R lavaan package was used to generate a latent variable measurement model that related each of the PCNB cognitive measures to the latent cognitive domain factors. The model was identified using the reference loading approach (29). Finally, within the ESEM model, each latent cognitive domain was regressed onto ACB scores. Models were fitted to the data using maximum likelihood estimation. Model fit was evaluated using root mean square error of approximation (RMSEA) and comparative fit index (CFI) metrics. RMSEA values less than 0.06 combined with CFI values greater than 0.95 are considered evidence of good model fit (30).

As shown in Figure 1, the average ACB score for study subjects was 3.8 (SD=2.9, range, 0–20). Most of the anticholinergic medication burden was attributable to antipsychotics (mean=2.1, SD=1.6), followed by traditional anticholinergics (benztropine, diphenhydramine, and trihexyphenidyl; mean=0.7, SD=1.3), antidepressants (mean=0.5, SD=0.8), mood stabilizers (mean=0.4, SD=1.0), and benzodiazepines (mean=0.1, SD=0.3). Among all study participants, 113 (10.1%) were not on antipsychotic therapy. Of those, 81 participants (7.2% of all included participants; 72% of those not on any antipsychotic therapy) had a medication regimen with an ACB score of 0.

The majority of schizophrenia patients were on antipsychotic monotherapy, with second-generation agents outnumbering first-generation agents by a ratio of 9:1 (see Table 1); approximately one out of five patients were on antipsychotic polypharmacy. Of those on monotherapy, ACB scores tended to be higher among those on first-generation compared with second-generation antipsychotics (total mean ACB score, 4.3 [SD=2.5] for first-generation agents and 3.2 [SD=2.0] for second-generation agents; F=20.4, p<0.001), driven by the use of traditional anticholinergics (mean ACB scores due to anticholinergics were 1.4 [SD=1.7] for patients taking first-generation antipsychotics and 0.4 [SD=1.1] for those taking second-generation antipsychotics; F=41.6, p<0.001). No other ACB score differences attributable to medication classes were detected when comparing between patients on first-generation and second-generation antipsychotic monotherapy (all other F values <1, p values >0.5). Notably, there was no difference in ACB score attributable to antipsychotics between those on first- or second-generation antipsychotic monotherapy (mean ACB scores were 2.0 [SD=1.0] for first-generation antipsychotics and 1.8 [SD=1.0] for second-generation antipsychotics; F=1.5, p=0.21). The contribution of different medication classes to cumulative ACB score over the range of ACB scores observed in participants is illustrated in Figure 2.

As shown in Figure 3, ACB score was negatively associated with PCNB performance across all efficiency domains; higher anticholinergic medication burden was associated with worse cognitive performance. The PCNB global cognitive averages for none, low, average, high, and very high anticholinergic burdens were, respectively (in z values), −0.51, −0.70, −0.85, −0.96, and −1.15. Anticholinergic medication burden level was also associated with PCNB motor speed (F=9.3, p<0.001) and motor praxis (F=11.2, p<0.001). Similar effects were seen on 15 of 16 additional cognitive tests, as shown in Table S2 in the online supplement.

Notably, when controlling for total ACB score, there were no significant differences in cognition across all PCNB domains among schizophrenia patients on second-generation antipsychotic monotherapy, first-generation antipsychotic monotherapy, or polypharmacy (F=1.4, p=0.256). Among those on antipsychotic monotherapy, after controlling for total ACB score, there were no significant effects of individual antipsychotics in each class across all PCNB cognitive domains (second-generation antipsychotics, F=1.6, p=0.093; first-generation antipsychotics, F=0.99, p=0.454).

The negative association between ACB score and cognitive performance remained significant after individually controlling for antipsychotic dosage, cigarettes smoked per day, positive symptoms, negative symptoms, duration of illness, age, and number of past hospitalizations at time of enrollment in COGS-2 (see Table S3 in the online supplement).

Psychometric analyses identified four latent cognitive domains from the PCNB with factor structures consistent with previous studies in schizophrenia patients using a similar version of the battery (24): memory, attention and control, executive and visuospatial, and motor. The ESEM model provided an excellent fit to the data (RMSEA=0.048; CFI=0.991). Parameter estimates are reported in Figure 4. The effect of ACB score on these factors was similar, suggesting that anticholinergic medication burden effects on cognition in schizophrenia are not specific to any of the domains examined.

This study should be considered in the context of several limitations. First, COGS-2 used a cross-sectional design that makes it impossible to disentangle cause-effect relationships between anticholinergic burden and cognition or to determine the longitudinal impact of anticholinergic burden on cognitive functioning in schizophrenia. Since participants were not randomized to treatment in this study, other confounding factors or proxies for disease severity may account for the association of ACB score and cognition. However, the present findings are fully compatible with results from recent large-scale longitudinal studies of anticholinergic medication burden in adults without schizophrenia.

Second, as noted above, the impact of anticholinergic medication burden on cognition in schizophrenia is likely to be underestimated in this study given that schizophrenia patients who had significant medical conditions were not recruited for COGS-2. Moreover, while we excluded those with major medical issues that could confound results, treatments of more “minor” medical issues were not reported (e.g., allergies, gastroesophageal reflux disease, insomnia). To the degree that intermittent medications used for treating these medical issues may have anticholinergic properties, they may have contributed to unappreciated anticholinergic burden in our sample; intermittent use of over-the-counter medications with anticholinergic properties at the time of testing could not be reliably disambiguated. Similarly, we did not formally assess for mood or motor symptoms with instruments validated for these specific illness domains. We acknowledge that there are likely independent effects of mood (particularly for those with schizoaffective disorder, depressed type) as well of the full range of extrapyramidal side effects on cognitive performance and study outcomes, which must be carefully characterized in future studies.

Third, the ACB scale, along with other similar scales, is limited by a failure to account for dose as well as previous or longitudinal exposure to medications with anticholinergic properties, and our modified ACB scale included consensus values for several medications and a novel binning strategy, which, while anchored by empirical data from previous studies, could have affected results. While further work is needed to replicate, clarify, and delineate dose and time-course effects of anticholinergic burden and cognition for schizophrenia patients, our modified ACB rating scale and other similar scales offer other advantages: they are easily deployable in routine clinical settings and can be readily incorporated into electronic medication records, as well as future or retrospective research studies.

Fourth, as with many large-scale studies investigating effects of psychotropic medications in patients with chronic psychotic disorders, it is possible that medication nonadherence may have affected our results. This is an ongoing challenge in the field, and future studies may benefit from more active methods of medication adherence surveillance.

Lastly, the COGS-2 cohort largely consisted of schizophrenia participants with chronic illness. The effects described in our analyses may not hold for clinical high risk, prodromal, or early illness populations; similarly, the relationship between ACB score and cognitive functioning may be different during an acute decompensation or shortly after remission. Further studies are also needed to determine the longitudinal course and durability of anticholinergic burden–cognition relationships.

Anticholinergic medication burden associated with psychotropic medications in schizophrenia is substantial, common, and conferred by multiple medication classes, including antipsychotics. Cumulative anticholinergic medication burden should be considered when prescribing medications (psychotropic as well as nonpsychotropic) for patients with schizophrenia. Efforts to limit or avoid excessive anticholinergic medication burden—regardless of source—may have a beneficial impact on cognitive outcomes in schizophrenia. Algorithms for anticholinergic burden calculations may be deployable within electronic medical record systems to guide providers in medication choices. When deprescribing or avoiding a high–anticholinergic burden regimen may not be clinically prudent, adjuvant treatment that can reduce the negative cognitive impact of anticholinergic medication burden may be helpful. For example, we recently found that a computerized cognitive training intervention blunted some anticholinergic burden–associated cognitive deterioration in chronic schizophrenia patients in residential inpatient care (23). Studies are needed to assess the longitudinal cognitive impact of anticholinergic burden and thereby further guide and optimize prescribing practices.