Minor Impact of Multiple Psychiatric Symptoms on Quality of Life (EQ5D) in Psychogeriatric Patients: A Clinical-Empirical Structural Modeling Approach Hugo J. Duivenvoorden, Ph.D., Ton Bakker, M.D., Ph.D.

Background: The relationship of clinical variables for cognitive functioning and other variables such as multiple neuropsychiatric symptoms and quality of life are usually analyzed bivariately and multivariately. In randomized controlled trials analyses are mostly performed on individual outcome variables. To unravel interdependencies of determinants and outcome variables, Structural Equation Modeling (SEM) was applied. Methods: Using SEM, we explored interdependencies of clinical determinants (MMSE, BI, and NPI-sum severity) and quality of life (EQ5D) in psychogeriatric patients (including dementia) suffering from cognitive problems and multiple neuropsychiatric symptoms. Results: MMSE and BI showed direct and indirect impact on EQ5D, but the association with NPI-sum severity was minor. The identified model showed that R2 of EQ5D varied from 0.21 to 0.48. Discussion: This clinical-empirical study showed that SEM could be utilized to unravel and identify a model of complex direct and indirect effects of MMSE and BI on EQ5D. In relation to NPI-sum severity, however, the validity of EQ5D seemed insufficient in psychogeriatric patients. Consequently, the cost-benefit analyses and cost-effectiveness analyses using quality-adjusted life-year measures of an intervention in psychogeriatric patients with multiple neuropsychiatric symptoms, based on EQ5D, are considered questionable. Construction of a quality of life instrument that is more sensitive with regard to multiple neuropsychiatric symptoms in psychogeriatric patients is highly recommended. (Am J Geriatr Psychiatry 2014; -:-e-) Key Words: Dementia, cognitive impairment, quality of life, EQ5D, cost-benefit, neuropsychiatric symptoms, modeling

I

n psychogeriatric patients who suffer from cognitive impairment or dementia, there is an 80% prevalence of two or more neuropsychiatric symptoms—for

example, depression, anxiety, paranoia, or aggression.1e5 Multiple psychiatric symptoms (MPS), belonging to the top three problems experienced by

Received July 26, 2013; revised February 25, 2014; accepted February 25, 2014. From the Erasmus University Medical Centre (HJD), Erasmus Rotterdam, the Netherlands; and Argos Care Institution (HJD, TB), Rotterdam, the Netherlands. Send correspondence and reprint requests to Ton Bakker, M.D., Ph.D., Argos Zorggroep, Postbox 4023, 3102 GA Schiedam, The Netherlands. e-mail: [email protected] Ó 2014 American Association for Geriatric Psychiatry http://dx.doi.org/10.1016/j.jagp.2014.02.008

Am J Geriatr Psychiatry

-:-, -

2014

1

Multiple Psychiatric Symptoms on Quality of Life, Modeled psychogeriatric patients and their caregivers,6 have negative effects on cognitive functioning and quality of life. Psychotropic drugs are widely used to treat MPS of dementia patients despite of their limited effects and potentially harmful side effects (e.g., (a)typical antipsychotics1,2,7). Psychotherapeutic interventions focusing on the MPS of psychogeriatric patients (including dementia) are complex due to their magnitude and the coexistence of cognitive, somatic, and social problems (e.g., loneliness).8,9 We developed an integrative psychotherapeutic nursing home program, Integrative Reactivation and Rehabilitation (IRR). IRR as compared to usual care (UC) was evaluated in a large-scale randomized clinical trial.10 This evaluation also included an economic objective in that the cost-benefit of IRR was assessed in terms of quality-adjusted life-year measures (QALYs), based on Euroqol5D (EQ5D) scores. In addition, the cost-effectiveness of IRR in terms of incremental cost-effectiveness ratios (ICERs; the ratio of the change in costs and change in benefits of a therapeutic intervention) was compared to UC. In this study we found a remarkable difference in results between QALYs and ICERs. The ICERs of clinically relevant outcomes (i.e., severity of psychiatric symptoms of psychogeriatric patients, caregiver burden, and competence) were clearly in favor for IRR, with relatively large numbers of improved patients (0.5 standard deviation [SD]10,11) and low numbers needed to treat (NNT ¼ 4 for Neuropsychiatric Inventory). In comparison, the NNT for donezepil is 10, for memantine the NNT is 3e8, and for cognitive behavior therapy the NNT is 5e10.12 In contrast, the QALYs of patients were almost equal between IRR and usual care. QALY calculation is based on the scores of EQ5D, a measure of health outcome. One plausible explanation of the low impact of MPS on quality of life of psychogeriatric patients suffering from multiple psychiatric symptoms is that EQ5D was relatively irresponsive to changes in MPS of quality of life. The mean differences on EQ5D between the conditions was small (0.04), with relatively small numbers of clinically relevant (0.5 SD) improved patients.10,11,13 This corresponds with the findings of Ballard et al.14 and Katona et al.,15 showing that clinically relevant improvement on behavioral and psychological symptoms of dementia had only small effects on regular quality of life measurements. This may mean

2

that cost-benefit studies in psychogeriatrics need to be reconsidered.16 One way to explore this finding further is to use appropriate statistical methods that consider the complex interplays of determinants and outcome variables. Usually, in randomized controlled trials the cost-effectiveness analyses are performed on the individual outcome variables. Likewise, the relationship of clinical variables for cognitive functioning, MPS, and other variables (e.g., caregiver factors) to quality of life are analyzed bivariately and sometimes multivariately.17,18 It is of clinical interest to unravel the relationship of EQ5D as a measure of quality of life to clinical relevant outcome variables (i.e., multiple psychiatric and cognitive symptoms), especially in the context of cost-benefit analyses in psychogeriatrics. Structural equation modeling (SEM) is an appropriate statistical method that enables solving the often complex interdependencies of the determinants and outcome variables.19e21 SEM has been applied in several research fields, but it is rarely used in psychiatric and geriatric research. We used SEM to describe the relationship of EQ5D to the clinical relevant determinants. Our objective was to explore the interrelationship of the Mini Mental State Exam (MMSE), the Barthel Index (BI), and Neuropsychiatric Inventory (NPI) sum severity and EQ5D across time, before and after IRR and UC intervention.

METHODS The details of the study have been described elsewhere.10,16 We provide a summary here. Patients The participants were recruited from the urban region of Nieuwe Waterweg Noord, near Rotterdam in the Netherlands. Inclusion criteria were: 1) a DSM-IV classification for dementia, amnestic disorders, or other cognitive disorders; 2) age 65 years or older; 3) cognitive functioning: MMSE between 18 and 27 as well as a BI between 5 and 19; 4) psychiatric symptoms: NPI ¼ 3 or more symptoms; and 5) informed consent. Exclusion criteria were: 1) delirium; 2) life-threatening somatic co-morbidity; 3) active coercive admission regime (according to psychiatric legislation); and 4) insufficient command of the Dutch language.

Am J Geriatr Psychiatry

-:-, -

2014

Duivenvoorden and Bakker

FIGURE 1. Flowchart of IRR patient protocol and percentages of patients receiving treatment components per phase.

Assessments The following demographic data were collected from the patient and the caregiver: sex, age, marital status, family relationship, domicile, and level of education. The severity of MPS assessed by number (ranging from 0 to 12)  frequency (ranging from 1 [sometimes] to 4 [very often])  severity (ranging from 1 [light] to 3 [severe]) using the NPI (12-item version).22 The NPI was administered to the caregiver. To asses

Am J Geriatr Psychiatry

-:-, -

2014

the cognitive functioning of the patient, memory was measured with the MMSE (0e30; 30 ¼ normal)23 and self-care with the BI (0e20; 20 ¼ normal).24 We include EQ5D as a measure of quality of life (EQ5D: 0.59 to 1.0; 1.0 ¼ optimal).25 Design The study was an open randomized controlled trial with a parallel group design and was performed

3

Multiple Psychiatric Symptoms on Quality of Life, Modeled from 2001 until 2006. Assessments were conducted simultaneously in both groups at three measurement points: T0 (within two weeks after inclusion), T1 (at the end of the intervention, about 3 months after inclusion) and T2 (follow-up, 6 months after the end of the intervention). The study protocol was approved by the Medical Ethics Committee of the Erasmus University Medical Centre. Intervention The duration of the IRR program was 13 weeks, with clinical admission to a separate 15-bed specialized unit in a psychiatric-skilled nursing home. IRR is meant as a short-stay reactivation and rehabilitation unit in addition to the usual multidisciplinary nursing home care, including psychotropic drug treatment. IRR consisted of both integrative psychotherapeutic interventions to treat multiple psychiatric symptoms of the patient and family therapy for the caregiver (Fig. 1). UC consisted of a relatively high level of multidisciplinary nursing home care provided by a multidisciplinary team in various care settings. Statistical Analyses Basic analyses. Fisher’s exact tests were used to estimate differences between IRR and UC on counts such as the number of deaths. Student’s t test for unpaired samples was used to test for differences between IRR and UC on continuous data. To evaluate the effect of IRR compared with UC, the mean differences on the continuous outcome variables were calculated over time. Advanced analyses. To unravel the interrelationships of the outcome variables and its determinants, we used SEM, a powerful statistical tool for pathanalysis, using a maximum likelihood estimation, namely MLR.20 We explored several possibilities to identify, test, and estimate models. Although there are no absolute standards for the relation between sample size and model complexity, a desirable goal is to have a minimal participant/ parameter ratio of 10:1.26 In order to restrict the number of parameters to be estimated because of the relatively small sample size, we have refrained from analyses of cross-regressions across time (lags 1 and 2). In addition, we have refrained from separate analyses of the two interventions. Nevertheless, the

4

baseline measurements of the clinical variables and EQ5D were adjusted for type of intervention. Four rival models were evaluated. First of all, we explored whether autoregressions of the variables from baseline (T0) to follow-up (T2) were required for adequate fit of the statistical model; and secondly, whether paths drawn from clinical variables to EQ5D were more plausible, or rather paths from EQ5D to the clinical variables, although we believe that the latter is clinically counterintuitive. We used standardized regression coefficients as effect estimates (theoretical range: 1.0 to þ1.0). Effect sizes can be reasonably estimated in combination with tests of significance, which also take into account the sample size. In addition, the percentages of explained variances of the outcome variables across time by the determinants are presented. Analysis strategy. The analysis strategy comprised two steps. Step 1. Mutual associations between the clinical determinants (MMSE, BI and NPI-sum severity) were analyzed. Step 2. We explored whether the clinical determinants were associated with the quality of life variable at T0, T1, and T2. Autoregressions across time (lag 1: between times 0 and 1 as well as between times 1 and 2) were estimated for the quality of life outcome variable EQ5D and the three clinical determinants (MMSE, BI, and NPI-sum severity). We also explored whether the autoregressions of lag 2 (between times 0 and 2) of the three clinical determinants and EQ5D were required to adequately fit the statistical model. Simultaneously, the cross-regressions of assessments EQ5D on the three clinical determinants MMSE, BI, and NPI-sum severity at the same time were estimated (lag 0). Regarding missing data, the maximum likelihood estimation under MAR was used. MAR means that missingness was a function of the observed values of both the outcome variable and the clinical determinants. For each model, we evaluated the model fit by measures of overall fit and detailed inspection of fit (fitted and standardized residuals and modification indices), as well as by examining the individual parameter estimates. The following performance measures were used: 1) c2-statistic for model fit (low and non-significant values of the c2 are desired);26 2) c2/degrees of freedom-ratio (a value 1.0 indicate an overidentification); 5) root mean square error of approximation (RMSEA: a value

Minor impact of multiple psychiatric symptoms on quality of life (EQ5D) in psychogeriatric patients: a clinical-empirical structural modeling approach.

The relationship of clinical variables for cognitive functioning and other variables such as multiple neuropsychiatric symptoms and quality of life ar...
1MB Sizes 0 Downloads 3 Views