**2.3 Discussion**

*Aging - Life Span and Life Expectancy*

Male

No prescriptions

**Table 8.**

**Figure 6.**

**16**

*2.2.5 Additional analyses*

residents with only PRN prescriptions are also unsurprising. The new ground breaking findings, however, concern residents with both daily *and* PRN prescriptions. This combination significantly augments mortality from levels that typify daily prescription and no prescription, but significantly attenuates levels that typify only PRN prescriptions. **Figure 6** shows that the respective mortality levels are independent of mortality risk, as estimated by categories of CHESS scores. Consequently, we infer that PRN prescription may overturn protective effects associated with daily prescriptions, whereas the latter may reduce the deleterious effects of PRN prescribing.

**Model term Odds ratio Sig. 95% confidence interval**

Intercept 0.21 0.000 0.19 0.23 Female 0.67 0.000 0.63 0.72

Age at assessment 1.03 0.000 1.02 1.03 Activities of daily living 1.36 0.000 1.32 1.40 Cognitive performance 0.97 0.007 0.95 0.99 CHESS scale 1.58 0.000 1.53 1.62 Only daily prescriptions 0.76 0.000 0.69 0.84 Only PRN prescriptions 1.42 0.000 1.21 1.66 Daily and PRN prescriptions 1.17 0.025 1.02 1.34

*Fixed effect odds ratios for mortality within 90 days of the final assessment.*

*Proportionate mortality by prescription regimens and CHESS categories.*

**Lower Upper**

Because of the statistical significance of LTCH as a random effect in all the preceding analyses, we report here on additional analyses that explore whether

This chapter charts a journey of scientific investigation with the following milestones:


The journey began with concern about excess mortality with the use of antipsychotic medication but ends with evidence for minor protective effects of daily prescription. How do we account for this discrepancy? Consider, first, the quality of measurement and, second, the appropriateness of the analyses. We have confidence in the data quality and psychometric adequacy (i.e., reliability and validity) of the RAI 2.0 measures. For nearly 30 years, efforts by researchers from across the world tried to ensure good measurement quality. A recent example is research on 15 years of archived data in Canada. Those authors report favorable outcomes, concluding that the RAI 2.0 provides a "robust, high quality data source" ([28], p. 27). If our data possess unrecognized error in measurement, similar error likely contaminates findings from many hundred referred publications that rely on RAI 2.0 data.

Second, we consider the technical details of our methodology to be appropriate. The decision to restrict analysis to data from new admissions ensures that the findings from over 20,000 residents provide 1-year incidence rates (rather than prevalence rates) for mortality. Such restriction also ensures that the data from all residents begin with comparable intake assessments. Earlier in this chapter, we provide conceptual reasons to justify the appropriateness of GLMM analysis. In every analysis, findings of significant random effects for LTCH provide empirical justification. Our wish for the future is that more researchers adopt similar modeling in order to avoid incorrect assumptions that measurement error is uncorrelated in clustered data.

We also ought to mention here that the statistical models show reasonable data fit in every analysis. The final model in **Table 8** provides an example. Compared to a simplified model that includes only demographics as fixed effects, the proportions of correct classification in the full model improve by 16.3% for predicted mortality (i.e., from 50 to 66.3%) and by 2.4% for the prediction of nonmortality (i.e., from 82.1 to 84.5%). These gains are fairly impressive given the lopsided skew in mortality data.

We conclude from the preceding is that our findings are trustworthy. How then can we account for findings from other studies of excessive mortality among elderly people prescribed with antipsychotic medication? We offer three possible interpretations. The first is a failure by many or most such studies to distinguish between daily and PRN prescribing. The second is a failure to provide information about PRN prescription of other psychotropic medications. Because our findings suggest associations between augmented mortality and PRN prescription of *any* psychotropic, each of the preceding concerns anticipates overestimation of mortality.

The third interpretation relates to adherence. Anyone with work experience in LTCH settings knows that adherence to drug regimens is problematic. Residents that put pills into their mouths do not necessarily swallow them. Instead, some residents chose to hide their pills, others throw them away. Previous findings show that nonadherence to antidepressant medication has unfavorable implications for mortality [19]. Findings from psychiatry show frequent low adherence to antipsychotic medication [29]. That study reports rates of nonadherence to antipsychotic medication by psychiatric patients as high as 40% within a year. Because low adherence results in intermittent use of antipsychotic medication, the consequences in LTCH residents could well include excessive mortality.

Our scientific journey progressed past milestones that indicate consistency in findings with alternative indices of mortality and control of confounding variables. At the end of the journey, the metaphorical villain is no longer daily prescription of psychotropic medication but PRN prescription. Although relatively few residents receive PRN prescriptions on a per drug basis, one resident in eight (12.5%) has at least one such prescription. Although residents prescribed PRN medication have high mortality risk, as measured by the CHESS, **Figure 6** shows elevated mortality rates at every level of mortality risk.

Issues that arise from our research include reasons for the high mortality associated with PRN prescriptions and implications for health care practice and policy. However, before such discussion, here are some words of caution.

The present research design is correlational rather that experimental. Only the latter assigns individuals randomly to the various conditions. Despite agreement that correlational designs provide less-compelling evidence about causality than experimental designs, the majority of studies on the effects of psychotropic medications are correlational. For example, most recent studies of the effects of antipsychotic medication on mortality have correlation designs that compare residents prescribed a given drug with residents without prescription or prescribed an alternative medication [6]. However, even in experimental studies, which are invariably of short duration [5], any differences in mortality between conditions do not necessarily imply direct causation. Such effects may arise because of changes

**19**

**3. Conclusion**

mortality associated with daily prescription.

*Psychotropic Medication Use and Mortality in Long-Term Care Residents*

controlled trials evaluate the relationships we report here.

which tackle an issue hitherto neglected in health care practice and policy.

PRN prescriptions of psychotropic medications have more aversive effects on the mortality of LTCH residents than daily prescriptions. These findings apply to all types of psychotropic medication after control for major confounding variables. The findings also suggest that daily prescription may ameliorate effects on mortality associated the PRN prescriptions, whereas PRN prescriptions exacerbate effects on

to health stability or changes in likelihood of adverse health events. We make these

Efforts within epidemiology to facilitate appropriate causal inferences include procedures to identify *confounding by indication*, which "refers to a situation where patient characteristics, rather than the intervention, are independent predictors of outcome" ([30], p. 841). The present analyses encompass procedures used to evaluate confounding by indication, which include, but are not limited to, the following. Covariate adjustment in which potential confounders are added to the list of fixed effects in modeling (e.g., age, gender, and functional measures). "Propensity" scores were used, which relate to the likelihood of intervention (e.g., high CHESS scores increase the likelihood of PRN prescription). The use of instrumental variables presumed to substitute for unmeasured confounders (e.g., LTCH and properties of LTCH as random and fixed effect variables, respectively). Although other research reports differences in causal direction depending on the kind of analysis [30], the present findings indicate directional consistency in relationships between mortality and PRN prescribing before (**Figures 1**–**5**) and after adjustment of variables from all three classes of confound.

However, our analyses clearly omit control for unmeasured sources of confounding. Also, our research lacks the rigor that random assignment brings to experimental forms of design. Consequently, we strongly recommend that future randomized

Our cautious interpretation of effects on mortality associated with daily and PRN prescription of psychotropic medication rests heavily on findings from the final analysis. These findings indicate that PRN prescription overturns any protective effects associated with daily prescription, whereas daily prescription lowers adverse consequences of PRN prescribing. These findings suggest a mutually antagonistic relationship between daily and PRN prescriptions with respect to their effects on mortality. The clinical rationale for daily prescription is the alleviation of disturbance to a resident's equilibrium, as exemplified by aggression, depression, pain, anxiety and insomnia. Successful treatment results in regained equilibrium after the resident adapts to regular ingestion of the drug. Conversely, the reasoning behind PRN prescription is to alleviate disturbances to equilibrium thought to be temporary. However, irregularity of ingestion is antagonistic to adaptation and may exacerbate instability through toxicity rather than help restore equilibrium. This interpretation appears to be consistent with our findings. Practical implications for health care practices and policy are consistent with current trends. Such trends include the de-prescribing of psychotropic medications for older people, with such medications to be used only for short durations and as a last resort [5, 6]. Similarly, a recent rewrite of Medicaid Long Term Care rules in the USA states that LTCH residents must not be prescribed antipsychotic, antidepressant, anxiolytic or hypnotic medications unless necessary to treat a specific condition diagnosed and documented in the clinical record. The new rules also normally limit PRN prescriptions to 14 days. Any extension requires an attending physician or prescribing practitioner to document the rationale in the resident's medical record [31]. We especially applaud these latter rules,

points to warn against simplistic interpretation of complex phenomena.

*DOI: http://dx.doi.org/10.5772/intechopen.85971*

### *Psychotropic Medication Use and Mortality in Long-Term Care Residents DOI: http://dx.doi.org/10.5772/intechopen.85971*

*Aging - Life Span and Life Expectancy*

rates at every level of mortality risk.

Second, we consider the technical details of our methodology to be appropriate. The decision to restrict analysis to data from new admissions ensures that the findings from over 20,000 residents provide 1-year incidence rates (rather than prevalence rates) for mortality. Such restriction also ensures that the data from all residents begin with comparable intake assessments. Earlier in this chapter, we provide conceptual reasons to justify the appropriateness of GLMM analysis. In every analysis, findings of significant random effects for LTCH provide empirical justification. Our wish for the future is that more researchers adopt similar modeling in order to avoid incorrect

We also ought to mention here that the statistical models show reasonable data fit in every analysis. The final model in **Table 8** provides an example. Compared to a simplified model that includes only demographics as fixed effects, the proportions of correct classification in the full model improve by 16.3% for predicted mortality (i.e., from 50 to 66.3%) and by 2.4% for the prediction of nonmortality (i.e., from 82.1 to 84.5%). These gains are fairly impressive given the lopsided skew in mortality data. We conclude from the preceding is that our findings are trustworthy. How then can we account for findings from other studies of excessive mortality among elderly people prescribed with antipsychotic medication? We offer three possible interpretations. The first is a failure by many or most such studies to distinguish between daily and PRN prescribing. The second is a failure to provide information about PRN prescription of other psychotropic medications. Because our findings suggest associations between augmented mortality and PRN prescription of *any* psychotro-

assumptions that measurement error is uncorrelated in clustered data.

pic, each of the preceding concerns anticipates overestimation of mortality.

The third interpretation relates to adherence. Anyone with work experience in LTCH settings knows that adherence to drug regimens is problematic. Residents that put pills into their mouths do not necessarily swallow them. Instead, some residents chose to hide their pills, others throw them away. Previous findings show that nonadherence to antidepressant medication has unfavorable implications for mortality [19]. Findings from psychiatry show frequent low adherence to antipsychotic medication [29]. That study reports rates of nonadherence to antipsychotic medication by psychiatric patients as high as 40% within a year. Because low adherence results in intermittent use of antipsychotic medication, the consequences in LTCH residents could well include excessive mortality. Our scientific journey progressed past milestones that indicate consistency in findings with alternative indices of mortality and control of confounding variables. At the end of the journey, the metaphorical villain is no longer daily prescription of psychotropic medication but PRN prescription. Although relatively few residents receive PRN prescriptions on a per drug basis, one resident in eight (12.5%) has at least one such prescription. Although residents prescribed PRN medication have high mortality risk, as measured by the CHESS, **Figure 6** shows elevated mortality

Issues that arise from our research include reasons for the high mortality associated with PRN prescriptions and implications for health care practice and policy.

The present research design is correlational rather that experimental. Only the latter assigns individuals randomly to the various conditions. Despite agreement that correlational designs provide less-compelling evidence about causality than experimental designs, the majority of studies on the effects of psychotropic medications are correlational. For example, most recent studies of the effects of antipsychotic medication on mortality have correlation designs that compare residents prescribed a given drug with residents without prescription or prescribed an alternative medication [6]. However, even in experimental studies, which are invariably of short duration [5], any differences in mortality between conditions do not necessarily imply direct causation. Such effects may arise because of changes

However, before such discussion, here are some words of caution.

**18**

to health stability or changes in likelihood of adverse health events. We make these points to warn against simplistic interpretation of complex phenomena.

Efforts within epidemiology to facilitate appropriate causal inferences include procedures to identify *confounding by indication*, which "refers to a situation where patient characteristics, rather than the intervention, are independent predictors of outcome" ([30], p. 841). The present analyses encompass procedures used to evaluate confounding by indication, which include, but are not limited to, the following. Covariate adjustment in which potential confounders are added to the list of fixed effects in modeling (e.g., age, gender, and functional measures). "Propensity" scores were used, which relate to the likelihood of intervention (e.g., high CHESS scores increase the likelihood of PRN prescription). The use of instrumental variables presumed to substitute for unmeasured confounders (e.g., LTCH and properties of LTCH as random and fixed effect variables, respectively). Although other research reports differences in causal direction depending on the kind of analysis [30], the present findings indicate directional consistency in relationships between mortality and PRN prescribing before (**Figures 1**–**5**) and after adjustment of variables from all three classes of confound.

However, our analyses clearly omit control for unmeasured sources of confounding. Also, our research lacks the rigor that random assignment brings to experimental forms of design. Consequently, we strongly recommend that future randomized controlled trials evaluate the relationships we report here.

Our cautious interpretation of effects on mortality associated with daily and PRN prescription of psychotropic medication rests heavily on findings from the final analysis. These findings indicate that PRN prescription overturns any protective effects associated with daily prescription, whereas daily prescription lowers adverse consequences of PRN prescribing. These findings suggest a mutually antagonistic relationship between daily and PRN prescriptions with respect to their effects on mortality. The clinical rationale for daily prescription is the alleviation of disturbance to a resident's equilibrium, as exemplified by aggression, depression, pain, anxiety and insomnia. Successful treatment results in regained equilibrium after the resident adapts to regular ingestion of the drug. Conversely, the reasoning behind PRN prescription is to alleviate disturbances to equilibrium thought to be temporary. However, irregularity of ingestion is antagonistic to adaptation and may exacerbate instability through toxicity rather than help restore equilibrium. This interpretation appears to be consistent with our findings. Practical implications for health care practices and policy are consistent with current trends. Such trends include the de-prescribing of psychotropic medications for older people, with such medications to be used only for short durations and as a last resort [5, 6]. Similarly, a recent rewrite of Medicaid Long Term Care rules in the USA states that LTCH residents must not be prescribed antipsychotic, antidepressant, anxiolytic or hypnotic medications unless necessary to treat a specific condition diagnosed and documented in the clinical record. The new rules also normally limit PRN prescriptions to 14 days. Any extension requires an attending physician or prescribing practitioner to document the rationale in the resident's medical record [31]. We especially applaud these latter rules, which tackle an issue hitherto neglected in health care practice and policy.
