Introduction
Despite advances in effective antiretroviral therapies (ART), 45% of people with HIV (PWH) continue to experience neurocognitive impairment (NCI) (Heaton et al.
2010; Wei et al.
2020). These deficits among PWH are associated with complications in everyday function (e.g., unemployment, and poor medication adherence) (Woods et al.
2008; Laverick et al.
2017; Marquine et al.
2018) making it important to identify factors that render PWH vulnerable to NCI. One factor may be elevated rates of medical and psychiatric conditions among PWH and the extensive use of non-ART medications to treat these conditions. Benzodiazepines, the most frequently prescribed medication for anxiety and sleep disorders, have been implicated in cognitive deficits in the general population. Although not consistently (Fastbom et al.
1998; Verdoux et al.
2005; Zhang et al.
2016; Nader and Gowing
2020), epidemiological studies suggest that benzodiazepine use is a risk factor for NCI and/or cognitive decline (Barker et al.
2004; Stewart
2005; Wright et al.
2009; Dell’Osso et al.
2015; Picton et al.
2018). Benzodiazepine use, particularly long-term use (at-least 6 months), is also associated with elevated dementia risk (Lagnaoui et al.
2002; Wu et al.
2009,
2011; Boeuf-Cazou et al.
2011; Billioti de Gage et al.
2012,
2014; Gallacher et al.
2012).
These adverse effects of benzodiazepines may be particularly concerning among individuals with health conditions characterized by neurobehavioral dysfunction such as HIV. Our group previously found that, among PWH, benzodiazepine users had a higher risk for NCI than benzodiazepine non-users even with adjustment for demographics, symptoms of affective distress, clinical comorbidities (e.g., hypertension and diabetes), and HIV-related disease factors (e.g., nadir CD4) (Saloner et al.
2019b). These adverse effects of benzodiazepine use were driven by the domains of processing speed, motor function, executive function, and memory suggesting effects of benzodiazepines on neurocognitive slowing and deficits in memory and higher-order cognitive capacities. These findings hold great public health significance given higher rates of benzodiazepine use among PWH (24%) compared to the general population (~ 13%) (Wixson and Brouwer
2014), likely due to the higher prevalence of anxiety and sleep disorders in PWH (Taibi
2013; Brandt et al.
2017).
However, an association between benzodiazepine use and NCI has not been found consistently in the general population (Fastbom et al.
1998; Billioti de Gage et al.
2014; Zhang et al.
2016; Nader and Gowing
2020). Discrepancies may be due to sample and methodological differences across studies and/or the potential moderating role of demographic or clinical characteristics. In our prior study, the benzodiazepine use and NCI link among PWH was not ubiquitous with 25% of benzodiazepine users not demonstrating NCI (Saloner et al.
2019b). Age potentially moderates the benzodiazepines and NCI relationship given that age-related changes in pharmacokinetics and pharmacodynamics can increase sensitivity to drug effects on the central nervous system (Kruse
1990; Hämmerlein et al.
1998; Turnheim
2003). Considering the aging demographic of PWH (Centers of Disease Control and Prevention
2020) and the elevated rates (Schouten et al.
2014) and earlier onset (De Francesco et al.
2020) of age-related comorbidities among PWH, a moderating role of age may be particularly important in the context of HIV. However, aging is a highly heterogeneous process, especially among PWH where different aging phenotypes are influenced by sociocultural, genetic, medication, comorbidities and inflammation factors (Stoff et al.
2017). Thus, clinical (e.g., medical burden and frailty index) and biological (e.g., telomere length and epigenetic clock) markers of aging are likely better indicators of the aging process than chronological age.
We explored the association between benzodiazepine and NCI among PWH by examining the moderating role of chronological age and medical burden, as an index of biological age, in a community-based sample of ART-treated, virally suppressed PWH. Medical burden was operationalized as an index representing the accumulation of multisystem health deficits that are both HIV-related and non-HIV-related (e.g., diabetes, anemia, and renal dysfunction). This medical burden index, in different adaptations, has been associated with cognition, everyday function, and mortality among PWH (Guaraldi et al.
2015; Oppenheim et al.
2018; Paolillo et al.
2019). We hypothesized that older chronological age and greater medical burden would separately exacerbate the adverse effect of benzodiazepine use on NCI, but medical burden would have a stronger moderating role than chronological age. In secondary analysis, we examined whether the moderating role of chronological age and/or medical burden on the benzodiazepine and NCI link was domain specific.
Discussion
In a virally suppressed sample of PWH, we extend previous finding of a relationship between benzodiazepine use and a higher likelihood of NCI by showing that this relationship depends on medical burden. Only among PWH with a medical burden index greater than 0.3 did benzodiazepine use relate to a higher likelihood of NCI. Conversely, the association between benzodiazepine use and cognitive function was not modified by chronological age. In domain-specific analyses, these results appeared to be driven by executive function, speed of information processing, motor and verbal fluency.
Medical burden, an indicator of biological age, but not chronological age moderated the relationship between benzodiazepine use and NCI in PWH. Although medical burden is often age-related, it is characteristically distinct from chronological age (Buchner and Wagner
1992; Fried et al.
2004). Similar to Oppenheim et al (
2018) (Oppenheim et al.
2018), we found a quadratic relationship whereby age positively related to medical burden only before age 63. Medical burden is postulated to reflect physiological or biological reserve more than chronological age in PWH given that age-related conditions and geriatric syndromes (e.g., frailty and dementia) are found in relatively young PWH (Brew et al.
2005; Soontornniyomkij et al.
2012; So-Armah et al.
2014; Greene et al.
2017), whereas a sizable proportion of older PWH experience no declines in cognitive or physical functioning (Saloner et al.
2019a; Fazeli et al.
2020). Therefore, the profile of health captured by the medical burden index may optimally represent biological age. Supporting this notion were findings from a study by Levine et al. (Levine et al.
2016) that created a measure of age acceleration based on the difference between biological aging, defined by DNA methylation levels, and chronological age in postmortem brain tissue of PWH. Greater age acceleration, but not chronological age, related to a higher likelihood of pre-mortem NCI. In the multisite CHARTER cohort, chronological age was strongly associated with poorer white matter integrity and less subcortical gray matter in PWH with severe comorbidity burden, yet this association was attenuated in PWH with minimal-to-moderate comorbidity burden (Saloner et al.
2019c).
When probing and benzodiazepine and NCI relationship, we found that both short- and long-acting BZDs related showed a relationship with higher odds of NCI that was of a similar magnitude although this relationship was only significant for short-acting benzodiazepines likely due to the larger cell size (n = 51 vs. 30). The consistency in findings between short- and long-acting benzodiazepines is similar to other studies reporting a relationship between benzodiazepine use and higher risk of dementia in the general population that did not differ between short- and long-acting substances (Gomm et al.
2016; Billioti de Gage et al.
2014); however, others have reported a stronger links between long-acting versus short-acting benzodiazepine use and cognitive impairment (Picton et al.
2018; Aldaz et al.
2021). Discrepancies are likely due to sample and methodological differences across studies and/or the potential confounding effects of other benzodiazepine use characteristics or clinical characteristics. Due to evidence in the literature of a stronger effect of benzodiazepine use on cognition with longer durations of use (Picton et al.
2018; Barker et al.
2004), it was unexpected that we did not observe a relationship between longer durations and odds of NCI among benzodiazepine users. However, this may be due to the limited variability in benzodiazepine use duration in our sample in that 70% of our sample reported a duration of 3–5 years. Given evidence that it is particularly long-term BZD use (at-least 1 year) that has adverse cognitive consequences (Nader and Gowing
2020; Barker et al.
2004; Crowe and Stranks
2018; Stewart
2005), the large proportion of long-term benzodiazepine use in our sample likely drove our findings.
Domain-specific analyses showed that the benzodiazepine use X medical burden interaction was evident in speed of information processing, verbal fluency and motor function. This is consistent with our prior study that also found an adverse effect of benzodiazepine use on speed of information processing and motor domains among PWH (Saloner et al.
2019b). A commonality among speed of information processing, motor function, and verbal fluency is that they are all speed-dependent and regulated by frontostriatal neural circuitry (Ances et al.
2012; Hakkers et al.
2017), which aligns with the theory that benzodiazepine causes psychomotor slowing (Rollings et al.
1994; Stewart
2005; Lader
2011). Frontostriatal circuity is also known to be disrupted in HIV (Du Plessis et al.
2014), which poses the question of potential compounding effects of benzodiazepine use and HIV infection on frontostriatal processing. In contrast, benzodiazepine use was independently associated with a higher likelihood of impairments in learning, recall and working memory/attention regardless of medical burden. This finding is also consistent with our prior findings that PWH who report benzodiazepine use and are positive for benzodiazepine in a toxicology test show working memory/attention and delayed recall impairments compared to PWH without evidence of benzodiazepine use (Saloner et al.
2019b). Similarly, a recent study from the Women’s Interagency HIV Study that reported an interaction between anxiolytic use (primarily benzodiazepines) and HIV serostatus on learning performance, whereby HIV-related learning deficits were only observed among those reporting anxiolytic use (Rubin et al.
2018). Our findings suggest that the adverse effects of benzodiazepine use on cognition may be more robust for the learning and memory domains given that it is observed consistently across studies and, in the current study, regardless of medical burden.
The underlying mechanisms and the temporal pattern of the benzodiazepine and NCI relationship are unclear. Insomnia and anxiety, the most common health conditions in which benzodiazepine are prescribed, can occur in the Alzheimer’s disease (AD) prodrome and, thus, the association may be an artifact of the use of benzodiazepines to treat early AD symptoms (Picton et al.
2018); however, suspected AD is currently rare among PWH (Morgello et al.
2018) given their shorter lifespan historically. Additionally, findings from longitudinal studies suggest a causative role of BZD on NCI including AD risk relating more strongly to long-term versus short-term benzodiazepine exposure (Billioti de Gage et al.
2014), and small improvements in memory performance among nursing home residents 4 weeks after benzodiazepine use was discontinued compared to residents that continued using benzodiazepines (Salzman et al.
1992).
There are potential reasons why benzodiazepine use is associated with NCI only in the context of high medical burden. It could suggest that the association between benzodiazepine use and NCI is more due to the cognitive effects of the underlying comorbidities rather than benzodiazepines themselves. However, we do not think this is likely because medical burden did not relate to NCI in benzodiazepine non-users as indicated by the nonsignificant medical burden term in our model including the medical burden X benzodiazepine interaction term. Another possibility is that multiple comorbidities could have cumulative effects on common underlying pathologies like chronic inflammation, hyperlipidemia and insulin resistance that weaken physiological reserve and heighten susceptibility to the effects of stressors; benzodiazepine in this case. HIV-related biological mechanisms and ART medications are associated with these common pathologies potentially making the cumulative effects of multimorbidity even more salient in HIV (D’Aversa et al.
2005; Deeks and Phillips
2009; Deeks
2011). Other measures of medical burden or physiological reserve including the Veterans Aging Cohort Study Index have been directly associated with NCI among PWH (Marquine et al.
2014,
2016; Zamudio-Rodríguez et al.
2018; Paolillo et al.
2020) suggesting that medical burden and benzodiazepine use may have compounding effects on NCI. Additionally, multimorbidity is typically coupled with polypharmacy, defined as more than 5 concurrent medications, which leads to higher risks of drug-drug and drug-disease interactions along with potentially cumulative toxicity. Although we covaried for antidepressant use and total number of medications, benzodiazepine may interact with specific medications resulting in compounding effects on brain health.
The clinical relevance of our findings is significant given the greater burden of comorbidities in PWH (Justice et al.
2012; Hosaka et al.
2019) and the higher rates of benzodiazepine use (Wixson and Brouwer
2014) versus the general population. Importantly, our sample of PWH were all on ART and virally suppressed indicating that the deleterious effects of combined benzodiazepine use and higher medical burden on NCI occur regardless of successfully treated HIV. Additionally, benzodiazepine use represents a modifiable risk factor. By identifying individuals at higher risk of benzodiazepine-related cognitive deficits, clinicians can prioritize alternative pharmaceutical or behavioral strategies to treating anxiety and insomnia in these patients. Lastly, our results suggest that medication use and medical burden may help explain some of the variability in cognitive function among PWH and more accurately characterize brain health than chronological age.
Our study has limitations. Assessment of benzodiazepine and some of the medical deficits were self-reported and thus subject to recall biases. Given that benzodiazepines are commonly prescribed for anxiety and sleep disorders, we would ideally adjust for these conditions. While were are able to adjust for GAD and anxiety/tension symptoms, we did not have data on sleep-related conditions. Our cross-sectional design precludes investigation of the temporal relationship between benzodiazepine use and NCI. Longitudinal studies are needed. Although it would be informative to compare individual types of benzodiazepine use in their relationship with NCI, we examined a combined group of all current benzodiazepine users in order to sustain adequate statistical power to test the interactive effects of benzodiazepine use, age, and medical burden. Given evidence in the general population that benzodiazepine effects on cognition are related to dose and duration (Barker et al.
2004; Crowe and Stranks
2018), the availability of these characteristics would have allowed us to more definitively test and characterize the benzodiazepine and NCI relationship. The absence of an HIV-uninfected comparison precluded us from determining whether our results are specific to the context of HIV. Study strengths include a large and neuropsychologically well-characterized sample of PWH and calculation of a comprehensive medical burden index of 28 health conditions that span multiple organ systems and show adverse health effects. Also, whereas studies typically contend with the confounding presence of comorbidities when assessing effects of medications on NCI, we directly modeled medical burden as a predictor/moderator.
Publisher's Note
Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.