The study identified 32,426 patients with PDP initiating atypical antipsychotic treatment in the overall PDP cohort before matching (Fig. 1); 4384 (13.5%) were pimavanserin initiators, and 28,042 (86.5%) were comparator atypical antipsychotic initiators. In the comparator group, quetiapine was the most frequently used antipsychotic (67.2%) (Online Resource 1, Table S3). Propensity scores were estimated and plotted by treatment group (Online Resource 1, Fig. S2). The PS overlap by treatment group suggested reasonable exchangeability between the pimavanserin group and the comparator group before matching. Matching retained 8762 individuals: 4381 (> 99.9%) of the pimavanserin group and 4381 (15.6%) of the comparator group (Fig. 1).

Attrition of patients with PDP initiating treatment with atypical antipsychotics after selection into the study cohort. DC District of Columbia, LTC/SNF long-term care/skilled nursing facility, PDP Parkinson’s disease psychosis, US United States. aPatients lacking 365 days of continuous fee-for-service coverage with Medicare Parts A, B, and/or D and/or who were not initially enrolled in Medicare based on entitlement due to age. bBipolar disorder, schizophrenia or schizoaffective disorder, major depressive disorder with psychotic symptoms

Selected characteristics of the overall PDP cohort before PS matching are shown in Table 1. Complete patient characteristics before matching are provided in Online Resource 1, Tables S4–S8. Before matching, key differences were observed between the treatment groups, with some comorbidities present less frequently in the pimavanserin group than in the comparator group (i.e., congestive heart failure [34.7% versus 45.4%], renal disease [20.2% versus 28.9%], use of ambulance/life support services [43.7% versus 59.1%], lower respiratory tract infections [11.8% versus 19.9%], and serious hospitalized infections [10.4% versus 19.2%]). With respect to healthcare utilization services, the pimavanserin group had a lower mean number of hospitalizations (0.5 versus 1.0) and emergency department visits (1.5 versus 2.1) than the comparator group. However, the pimavanserin group had a higher use of PD medications (95.2% versus 76.6%) and a higher mean number of unique days with a clinic visit (12.1 versus 10.7). The index medication prescriber was more frequently a neurologist for patients in the pimavanserin group than for those in the comparator group (61.7% versus 25.2%), while the prescriber being a primary care physician was more common for those in the comparator group than for those in the pimavanserin group (53.3% versus 25.7%). Dementia diagnoses were less common in the pimavanserin group than in the comparator group (69.2% versus 78.1%). The proportion of patients with a coronavirus disease 2019 (COVID-19) diagnosis was similar between the pimavanserin group and the comparator group in both hospitalized (0.8% versus 0.9%) and nonhospitalized (2.0% versus 2.2%) settings. After matching, all covariates were well balanced between treatment groups, as indicated by ASDs (Online Resource 1, Fig. S3). Since a small number of pimavanserin initiators failed to match (n = 3), characteristics of the matched cohorts cannot be displayed when cell size is < 11 patients owing to the Centers for Medicare and Medicaid Services privacy policies.

Patients were followed for a total of 18,015 person-years (mean [SD] per patient 0.6 [0.68] years) (Online Resource 1, Table S9). IRs and HRs of mortality are shown in Table 2. The crude IR per 100 person-years for mortality in the pimavanserin group (604 deaths) was 20.64 (95% CI 19.03–22.36) and in the comparator group (4914 deaths) was 32.57 (95% CI 31.66–33.49). The crude HR before matching for mortality among pimavanserin initiators compared with users of other atypical antipsychotics was 0.65 (95% CI 0.60–0.70). In the matched cohort, the IR per 100 person-years for mortality in the pimavanserin group (603 deaths) was 20.61 (95% CI 19.00–22.33) and in the comparator group (638 deaths) was 26.95 (95% CI 24.90–29.13). The matched HR for mortality among pimavanserin initiators compared with users of other atypical antipsychotics was 0.76 (95% CI 0.68–0.85).

To evaluate mortality risk over time since treatment initiation, the cumulative incidence of mortality was estimated and plotted in each treatment group, both before matching (Online Resource 1, Fig. S4) and in the matched cohort (Fig. 2). After matching, a reduced risk of mortality in the pimavanserin group was demonstrated by the cumulative incidence curves for at least the first year of follow-up. The curves began to converge after approximately 3.5 years of follow-up, but estimates were unstable owing to very small numbers of patients remaining in the cohort.

Cumulative incidence of mortality by time since atypical antipsychotic initiation among patients with PDP, after matching. PDP Parkinson’s disease psychosis

In addition, to describe changing risk more granularly during the first year of follow-up, time period-specific RRs and RDs were estimated in addition to the main HR (Fig. 3). On the relative scale, the lowest RRs were noted at day 90 (RR = 0.60; 95% CI 0.51–0.77) and day 180 (RR = 0.63; 95% CI 0.58–0.81) suggesting that pimavanserin initiators had a 37–40% lower risk of mortality compared with users of other atypical antipsychotics at these time points. At 1 year, although the relative mortality risk reduction was attenuated, pimavanserin initiators demonstrated a 22% lower risk of mortality compared with users of other atypical antipsychotics. On the absolute scale, the largest benefit of pimavanserin was seen from day 180 (RD = − 0.05; 95% CI − 0.02 to − 0.06) through day 365 (RD = −0.05; 95% CI − 0.01 to − 0.06) suggesting that pimavanserin initiators had a 5% lower absolute risk of mortality compared with users of other atypical antipsychotics. Presentation of risk on both relative and absolute scales allows for better interpretation of changing risk over time. The RR and RD estimates showed that compared with users of other atypical antipsychotics, pimavanserin initiators demonstrated a lower risk of mortality that was sustained throughout the first year of follow-up.

Matched RR and RD of mortality comparing patients with PDP using pimavanserin with patients using comparator atypical antipsychotics, overall and in specified follow-up periods. CI confidence interval, PDP Parkinson’s disease psychosis, RD risk difference, RR relative risk

The number of patients included in each subgroup analysis by treatment group before matching is provided in Online Resource 1, Table S10. In the subgroup analyses (i.e., sex, age groups, and dementia diagnosis), all covariates were well balanced after matching. The subgroup HR estimates for males, females, those aged 75–85 years, and dementia diagnosis were generally similar to the HR estimate for the overall PDP cohort (Fig. 4). The HR estimate was attenuated for the oldest age group (i.e., those aged ≥ 85 years).

Matched hazard ratios for mortality in patients with PDP who initiated treatment with pimavanserin compared with those initiating treatment with other atypical antipsychotics, main results and by subgroup. CI confidence interval, PDP Parkinson’s disease psychosis

The number of patients included in each sensitivity analysis by treatment group before matching is provided in Online Resource 1, Table S11. The matched HR estimates for each sensitivity analysis compared with those for the primary analysis are shown in Fig. 5. The HR estimates for all sensitivity analyses ranged from 0.72 (95% CI 0.61–0.84) to 0.86 (95% CI 0.77–0.96) and were similar to the HR for the overall primary PDP cohort (0.76). The HRs showed a reduced risk of mortality for the pimavanserin group compared with the comparator antipsychotic group in all sensitivity analyses. The lowest HR estimate was observed in the comparison of pimavanserin and non-quetiapine antipsychotics (HR = 0.72; 95% CI 0.61–0.84). The cumulative incidence curves for the pimavanserin versus quetiapine-alone comparator groups, both before and after matching, are shown in Online Resource 1, Figs. S5 and S6. The cumulative incidence curves for this analysis follow the same general pattern as the overall PDP analyses.

Matched hazard ratios of mortality in patients with PDP who initiated treatment with pimavanserin compared with those initiating treatment with other atypical antipsychotics, primary and sensitivity analyses. CI confidence interval, PDP Parkinson’s disease psychosis

Of the 32,426 patients identified in the overall PDP cohort, 8884 patients (27.4%) were residing in LTCs or SNFs on the index date: 921 patients were included in the pimavanserin group, and 7963 patients were included in the comparator group (Fig. 1). Propensity scores estimated separately for the LTC/SNF subcohort and plotted by treatment group (Online Resource 1, Fig. S5) suggested reasonable exchangeability between the pimavanserin group and the comparator group in the LTC/SNF subcohort. Matching retained 1810 individuals: 905 (98.3%) of the pimavanserin group and 905 (11.4%) of the comparator group (Fig. 1).

Select characteristics of the unmatched LTC/SNF subcohort by treatment group are shown in Table 3. Complete characteristics are provided in Online Resource 1, Tables S13–S17. Patients in the LTC/SNF subcohort generally had a higher comorbidity burden than those in the overall PDP cohort. Similar to the overall PDP cohort, within the LTC/SNF subcohort, fewer patients in the pimavanserin group than in the comparator group had severe comorbidities. However, the magnitude of difference between the pimavanserin and comparator groups was reduced in the LTC/SNF subcohort compared with the overall PDP cohort. After matching, all covariates were well balanced between treatment groups (Online Resource 1, Fig. S6).

Patients in the LTC/SNF subcohort were followed for a total of 4761 person-years (mean [SD] per patient, 0.5 [0.66] years) (Online Resource 1, Table S18). IRs and HRs of mortality for the LTC/SNF subcohort are shown in Table 4. In the LTC/SNF subcohort, the crude IR per 100 person-years for mortality in the pimavanserin group (186 deaths) was 36.07 (95% CI 31.07–41.64) and in the comparator group (1897 deaths) was 44.68 (95% CI 42.69–46.74). After matching, the IR per 100 person-years for mortality in the pimavanserin group (182 deaths) was 36.06 (95% CI 31.01–41.69) and in the comparator group (194 deaths) was 39.83 (95% CI 34.42–45.85). The matched HR for mortality in pimavanserin initiators compared with that for initiators of other atypical antipsychotics was 0.90 (95% CI 0.74–1.10). Incidence rates demonstrated that, regardless of treatment group, mortality was higher in the LTC/SNF subcohort than in the overall PDP cohort (pimavanserin group: IR = 36.06 versus 20.61; comparator group: IR = 39.83 versus 26.95).

After matching, there was no consistent difference in the risk of mortality between the pimavanserin group and the comparator antipsychotic group, as demonstrated by the cumulative incidence curves with overlapping CI bands (Fig. 6). Similarly, the time period-specific RR and RD estimates demonstrated little consistent difference between the groups over the first year of follow-up (Fig. 7).

Cumulative incidence of mortality by time since atypical antipsychotic initiation among patients with PDP residing in long-term care or skilled nursing facilities, after matching. PDP Parkinson’s disease psychosis

Matched RRs and RDs of mortality comparing patients with PDP residing in LTC/SNF using pimavanserin with patients using comparator atypical antipsychotics at specified follow-up periods. CI confidence interval, LTC/SNF long-term care/skilled nursing facility, PDP Parkinson’s disease psychosis, RD risk difference, RR relative risk

The number of patients in each subgroup by treatment group before matching are provided in Online Resource 1, Table S19. The HRs demonstrated no consistent differences in the risk of mortality in the pimavanserin group versus that in the comparator antipsychotic group within the subgroups of sex and dementia diagnosis (Online Resource 1, Fig. S8).

The number of patients included in each sensitivity analysis by treatment group before matching is provided in Online Resource 1, Table S20. The matched HRs for all sensitivity analyses (Online Resource 1, Fig. S9) ranged from 0.71 (95% CI 0.57–0.90) to 1.02 (95% CI 0.89–1.16), and were similar to the HRs for the LTC/SNF subcohort main results (HR = 0.90; 95% CI 0.74–1.10). The HRs showed a reduced risk of mortality in the pimavanserin group when the comparator antipsychotic group excluded quetiapine use. All other sensitivity analyses demonstrated no difference in the risk of mortality between the pimavanserin group and the comparator antipsychotic group. Sensitivity analysis of cumulative incidence curves for the pimavanserin versus quetiapine-alone comparator groups, both before and after matching, are shown in Online Resource 1, Figs. S10 and S11. The matched cumulative incidence curves for the sensitivity analysis were generally overlapping throughout follow-up, similar to those in the LTC/SNF subcohort analysis.