Study Model

We created a conceptual model adapted from the "Rapid Clinical Decision Making in Context" model [17], describing factors that may influence clinical decisions in nursing home residents in the setting of acute illness (AI). We included the following categories of factors as shown in Figure 1: provider knowledge and experience, patient clinical data, environmental context, patient expectations and communication. We hypothesized that influential factors vary during the episode with, for example, illness severity more influential at illness onset and hospital admission than at hospital discharge, and communication more influential at hospital admission and discharge than at illness onset. The main outcomes of acute illness in our model are those that are clinically important in nursing home patients such as function change, symptoms (pain, delirium, and medication withdrawal symptoms), and mortality.

Study Design

The study is a prospective observational study of nursing home residents in 2 nursing homes in metropolitan New York--Jewish Home Lifecare (JHL), New York, NY and the James J. Peters VA (JJP VA) Community Living Center (CLC) in Bronx, NY. These nursing homes were chosen because during acute illness episodes, residents needing hospital care are referred predominantly to the Mount Sinai Hospital for JHL residents, and to the James J Peters VA Medical Center for JJP VA CLC residents; thus we are able to follow residents at these hospitals, if necessary, during their acute illnesses to monitor withdrawal of CNS medications and to measure outcomes.

Eligibility Criteria

Residents are eligible to be enrolled in the study if they are receiving opioids, antidepressants, or antipsychotics on a routine basis for a duration of time considered to be a minimal therapeutic trial or in which tolerance develops. Because of the differences in pharmacokinetics and biologic effects among the 3 drug classes, a minimal therapeutic trial of 14, 30 and 7 days was selected for opioids, antidepressants, and antipsychotics respectively. We are including residents taking more than one medication in a class, or more than one class of medication if there are no known interactions between the medications. We are excluding residents receiving antidepressants who score fewer than 21 points on the Mini-Mental Status Examination [18] because the assessment of symptoms of depression, which is a main outcome, among those with more severe degrees of dementia, is less reliable. We are excluding residents who have an acute medical illness (defined below) at the time of screening and rescreening them for enrollment after it has resolved. Informed consent is obtained from residents or proxies. The study was approved by the Institutional Review Board at JHL and JJP VA Medical Center.

Acute Illness Surveillance and Points of Care Transition

Acute illness surveillance is performed twice weekly at the nursing home through communication with nursing home nursing staff and medical providers using established clinical criteria [19] for incipient cases. The clinical criteria for acute illness are meant to be sensitive for all acute medical problems experienced by nursing home residents and include clinical symptoms such as chest pain, dyspnea, diarrhea, acute change in mental status; clinical signs such as persistent increase or decrease in blood pressure, fever or hypothermia; and abnormalities in laboratory values such as a drop in hematocrit > 5 points with signs of acute bleeding. We follow patients for 14 days after illness onset, and if applicable, an additional 14 days each after hospital admission and hospital discharge. During each follow-up interval we assess the following signs and symptoms by patient and staff interview in the nursing home or in the hospital: pain, delirium, and withdrawal symptoms including gastrointestinal and cardiopulmonary signs and symptoms three times weekly, and mood and behavior once weekly, using validated instruments adapted for this population described below.

Measures

Baseline demographic information including age, gender, and race of each participant is collected. In addition, at baseline we collect information on chronic medical conditions and medication use through medical record abstraction, and on physical and cognitive function by interviews with patients, or proxies, and nursing home staff. Functional status is measured using items from the Minimum Data Set Activities of Daily Living Scale (MDS-ADL) adapted for interview with nursing home staff [20]. The MDS-ADL scale ranges from 0 to 6 with higher scores indicating poorer functional status. Cognitive function is measured using items from the Minimum Data Set Cognitive performance scale (MDS-CPS) [21] adapted for interview with nursing home staff and from the Mini Mental Status Examination (MMSE) [18]. Illness severity is measured using a physiologic measure, the Inpatient Physiologic Failure Score (IPFS) [22], which is adapted and validated for use in the elderly population. In the setting of acute illness, participants are assessed for delirium using the Confusion Assessment Method (CAM) [23].

Withdrawal symptoms from opiate withdrawal are measured using the Clinical Opiate Withdrawal Scale (COWS) [24], which contains measurements of several common signs and symptoms of withdrawal including tachycardia, sweating, restlessness, tremor, yawning, anxiety or irritability, gastrointestinal upset, pupil dilation, gooseflesh skin, running nose or tearing, and bone or joint aches. Scores on this scale vary from 0 to 48 with higher scores indicating more severe withdrawal. Withdrawal symptoms from antidepressants are measured using the Discontinuation Emergent Signs and Symptoms (DESS) scale [25]. We modified the scale to include only items within the proposed diagnostic criteria for Serotonin Reuptake Inhibitor syndrome [26], which includes the following symptoms--headache, insomnia, irritability, anxiety, fatigue, paresthesias, tremors, visual changes, dizziness, nausea, diarrhea, stomach cramps, chills, flushing and gait instability. We classify delirium as a separate complication from withdrawal, although we consider it consistent with withdrawal.

Relapse symptoms such as depression, psychosis and pain are measured using standardized instruments. Depression is measured using the Geriatric Depression Scale (GDS) [27], a 15-point scale for measuring depression in the elderly; a score of 5 or above indicates a positive screen for depression. Psychosis and disturbed behavior are measured using the Cohen Mansfield Agitation Inventory (CMAI) [28]. This scale contains 29 behaviors including verbal and physical behaviors, examples of which are repetitiveness, screaming, verbal aggression, and wandering. On the scale, these behaviors are measured on how frequently they are observed. Scores range from 29 to 103, with higher scores indicating more agitation. We used this scale as a measurement of the effect of antipsychotic cessation, because antipsychotic withdrawal has been associated with relapse symptoms [29], which can be reliably measured using this scale.

Pain is measured using a variety of measures based on whether the participant can reliably indicate their pain and severity. For participants who can indicate their pain, a modified Brief Pain Inventory (BPI) [30] is used, which measures the location and the severity of pain experienced currently and also in the past 24 hours. Participants are also asked to indicate how much their pain has interfered with their mood, sleep, walking ability and other activities. If the participant is unable to complete the BPI, a McGill Present Pain Intensity Scale [31] from 0 to 5 is attempted, and if unable to complete this, the participant is asked if they have pain currently (yes/no). If the participant is unable to provide answers regarding their pain, the checklist of nonverbal pain indicators (CNPI) [32] is used. Nursing staff are asked if they have observed any vocal complaints including verbal complaints and non-verbal sounds such as moans or groans, facial grimaces, restlessness, bracing, and rubbing during activity or rest.

Because symptoms of relapse and withdrawal can occur acutely and resolve in a short period of time, the pain, Clinical Opiate Withdrawal Scale (COWS), and Discontinuation Emergent Signs and Symptoms (DESS) scale are repeated three times a week during AI episodes to determine if withdrawal symptoms occurred. Functional status data are collected at the time of enrollment, subsequently at 3-month intervals and at the time of illness onset and 14 days after illness. Figure 2 outlines the timeline of our data collection during the study period.

Statistical Analysis

Baseline demographic characteristics and symptoms measured on scales described above are summarized using descriptive statistics. Characteristics of acute illness including illness type and severity, are described. Planned analyses include estimation of the effect of acute illness on functional decline by comparing magnitude of functional change during periods with acute illness with periods without acute illness. Additional analyses include a description of the pattern of interruption of CNS medications during the acute illness and, to examine predictors of CNS medication interruption, estimation of a multivariable logistic regression model using occurrence of interruption in CNS medication as dependent variable, and patient, illness, and provider characteristics as independent variables. We will select independent variables to include in the model based on whether a factor is associated with the outcome in univariate analysis.

To examine the impact of CNS medication interruption on patient outcomes, we plan to estimate multivariable logistic regression models with pain, depression and disturbed behavior at the moderate or severe level as dependent variables and interruption in CNS medications as independent variable. Covariates will include patient and illness characteristics outlined in Figure 1. Furthermore, we plan to describe the effect of the interruption of CNS medications on hospital use, illness duration and functional outcomes, using similar multivariable regression models.

We also plan to compare proportions of patients who developed withdrawal symptoms who had medication interruption to those who did not have medication interruption. For the sample size of 200 AI episodes, we estimated that 20% of them would have an interruption of CNS medications. Assuming that the proportion who would have withdrawal symptoms without interruption of CNS medications (due to other causes such as the acute illness itself) to be 15%, this will provide 83% power (with an alpha of 0.05) to detect an absolute difference of 25% in the proportion with interruption of CNS medications who develops withdrawal symptoms (with a proportion of 40%). To estimate the effect of acute illness on function, we collected functional status data at the time of enrollment, subsequently at 3 month intervals, and at the time of illness onset and 14 days after illness. We plan to use a linear mixed model to account for such repeated measurements.