Accumulating studies have assessed mortality risk associated with mood-stabilizers, the mainstay treatment for bipolar disorder (BD). However, existing data were mostly restricted to suicide risk, focused on lithium and valproate and rarely adequately adjusted for potential confounders. This study aimed to assess comparative mortality risk with all, natural and unnatural causes between lithium, valproate and three frequently prescribed second-generation antipsychotics (SGA), with adjustment for important confounders.

This population-based cohort study identified 8137 patients with first-diagnosed BD, who had exposed to lithium (n = 1028), valproate (n = 3580), olanzapine (n = 797), quetiapine (n = 1975) or risperidone (n = 757) between 2002 and 2018. Data were retrieved from territory-wide medical-record database of public healthcare services in Hong Kong. Propensity-score (PS)-weighting method was applied to optimize control for potential confounders including pre-existing chronic physical diseases, substance/alcohol use disorders and other psychotropic medications. PS-weighted Cox proportional-hazards regression was conducted to assess risk of all-, natural- and unnatural-cause mortality related to each mood-stabilizer, compared to lithium. Three sets of sensitivity analyses were conducted by restricting to patients with (i) length of cumulative exposure to specified mood-stabilizer ≥90 days and its medication possession ratio (MPR) ≥90%, (ii) MPR of specified mood-stabilizer ≥80% and MPR of other studied mood-stabilizers <20% and (iii) monotherapy.

Incidence rates of all-cause mortality per 1000 person-years were 5.9 (95% confidence interval [CI]: 4.5–7.6), 8.4 (7.4–9.5), 11.1 (8.3–14.9), 7.4 (6.0–9.2) and 12.0 (9.3–15.6) for lithium-, valproate-, olanzapine-, quetiapine- and risperidone-treated groups, respectively. BD patients treated with olanzapine (PS-weighted hazard ratio = 2.07 [95% CI: 1.33–3.22]) and risperidone (1.66 [1.08–2.55]) had significantly higher all-cause mortality rate than lithium-treated group. Olanzapine was associated with increased risk of natural-cause mortality (3.04 [1.54–6.00]) and risperidone was related to elevated risk of unnatural-cause mortality (3.33 [1.62–6.86]), relative to lithium. The association between olanzapine and increased natural-cause mortality rate was consistently affirmed in sensitivity analyses. Relationship between risperidone and elevated unnatural-cause mortality became non-significant in sensitivity analyses restricted to low MPR in other mood-stabilizers and monotherapy. Valproate- and lithium-treated groups did not show significant differences in all-, natural- or unnatural-cause mortality risk.

Our data showed that olanzapine and risperidone were associated with higher mortality risk than lithium, and further supported the clinical guidelines recommending lithium as the first-line mood-stabilizer for BD. Future research is required to further clarify comparative mortality risk associated with individual SGA agents to facilitate risk-benefit evaluation of alternative mood-stabilizers to minimize avoidable premature mortality in BD.

Approximately 6.5 million Americans ages 65 and older have Alzheimer’s disease and related dementias, a prevalence projected to triple by 2060. While subtle impairment in cognition and instrumental activities of daily living (IADLs) arises in the mild cognitive impairment (MCI) phase, early detection of these insidious changes is difficult to capture given limitations. Traditional IADL assessments administered infrequently are less sensitive to early MCI and not conducive to tracking subtle changes that precede significant declines. Continuous passive monitoring of IADLs using sensors and software in home environments is a promising alternative. The purpose of this study was to determine which remotely monitored IADLs best distinguish between MCI and normal cognition.

Participants were 65 years or older, independently community-dwelling, and had at least one daily medication and home internet access. Clinical assessments were performed at baseline. Electronic pillboxes (MedTracker) and computer software (Worktime) measured daily medication and computer habits using the Oregon Center for Aging and Technology (ORCATECH) platform. The Survey for Memory, Attention, and Reaction Time (SMART; Trail A, Trail B, and Stroop Tests) is a self-administered digital cognitive assessment that was deployed monthly. IADL data was aggregated for each participant at baseline (first 90 days) in each domain and various features developed for each. The receiver operating characteristic area under the curve (ROC-AUC) was calculated for each feature.

Traditional IADL Questionnaires.

At baseline, 103 participants (normal n = 59, Mage = 73.6±5.5; MCI n = 44, Mage = 76.0±6.1) completed three functional questionnaires (Functional Activities Questionnaire; Measurement of Everyday Cognition (ECog), both self-report and informant). The Informant ECog demonstrated the highest AUC (72% AUC, p = <.001).

Remotely monitored in-home IADLs and self-administered brief online cognitive test performance.

Eighty-four had medication data (normal n = 48, Mage = 73.2±5.4; MCI n = 36, Mage = 75.6±6.9). Four features related to pillbox-use frequency (73% AUC) and four features related to pillbox-use time (62% AUC) were developed. The discrepancy between self-reported frequency of use versus actual use was the most discriminating (67% AUC, p = .03).

Sixty-six had computer data (normal n = 38, Mage = 73.6±6.1; MCI n = 28, Mage = 76.6±6.8). Average usage time showed 64% AUC (p = .048) and usage variability showed 60% AUC (p = .18).

One hundred and two completed the SMART (normal n = 59, Mage = 73.6±5.5; MCI n = 43, Mage = 75.9±6.2). Eleven features related to survey completion time demonstrated 80% AUC in discriminating cognition. Eleven features related to the number of clicks during the survey demonstrated 70% AUC. Lastly, seven mouse movement features demonstrated 71% AUC.

Pillbox use frequency combined features and self-administered brief online cognitive test combined features (e.g., completion times, mouse cursor movements) have acceptable to excellent ability to discriminate between normal cognition and MCI and are relatively comparable to informant rated IADL questionnaires. General computer usage habits demonstrated lower discriminatory ability. Our approach has applied implications for detecting and tracking older adults’ declining cognition and function in real world contexts.

Subjective cognitive decline (SCD, i.e., perceived cognitive decline without neuropsychological deficits) is associated with Alzheimer’s disease pathology and increased risk for cognitive impairment but is heterogenous in etiology and has been linked to other factors including personality and depression. Mental wellbeing (i.e., the perception and functioning of social, emotional, and health-related aspects of one’s life) has been associated with subjective memory complaints, but its relationship with other subjective cognitive domains is poorly understood. Further characterizing the relationship between mental wellbeing and SCD could refine understanding of SCD and inform development of interventions that prevent progression to objective cognitive decline. This study aimed to describe the relationship between mental wellbeing and subjective decline in multiple cognitive domains and examine whether this relationship differs between older adults with normal cognition and those with mild cognitive impairment (MCI).

Community-dwelling older adults (normal: n = 58, Mage = 73.7±5.6; MCI: n = 43, Mage = 75.9±6.1) completed the Everyday Cognition scale, a validated self-report measure of SCD, and the RAND-36 Health Survey, a validated self-report measure of health-related quality of life which includes a mental wellbeing subscale. Spearman’s rank correlations were conducted between self-reported mental wellbeing and each self-reported cognitive domain (i.e., memory, language, visuospatial, and executive function) for the Normal Cognition and MCI groups.

Worse mental wellbeing was associated with worse subjective language and executive function in the normal group, rs(56) = -.42, p =.001; rs(56) = -.37, p =.005, but not for the MCI group, rs(41) = -.23, p =.15; rs(41) = -.12, p =.46. Worse mental wellbeing was associated with worse subjective visuospatial function in the MCI group, rs(41) = -.39, p =.009, but not in the normal group, rs(56) = -.11, p =.39. For both groups, worse mental wellbeing was associated with worse subjective memory, rs(56) = -.45, p < .001; rs(41) = -.37, p =.02. While this correlation was greater in the normal group, the difference was not significant (z = 0.38, p =.71).

These results suggest that perceptions of mental wellbeing are related to perceptions of cognitive decline in multiple domains, and that the specific domains involved differ between normal and MCI groups. The differential associations may mean perception of specific cognitive domains more strongly affect mental wellbeing, or mental wellbeing more acutely influences perception of those domains. The overall observed relationship between SCD and mental wellbeing may have several explanations: the impact of broader health perceptions may extend to cognitive perception, behavioral changes associated with poor wellbeing may reduce subjective cognitive function, or worse subjective cognitive function may lead to negative experiences of wellbeing. Future longitudinal investigation could inform causal inferences. The more limited associations between mental wellbeing and SCD among MCI individuals may point to the role of decreased self-awareness (due to cognitive impairment) precluding detection of subtle changes in cognition or wellbeing. This study highlights the importance of better understanding mental wellbeing in experiences of SCD in both normal and MCI older adults to improve cognitive and mental health outcomes.