Validating a Prognostic Model for 1-Year Mortality of Psychogeriatric Inpatients.

Background To validate a predictive scoring system for 1-year mortality among psychogeriatric inpatients admitted for acute psychiatric care. Methods Computerized data were extracted from the District Health Board Database for a university aliated general hospital. A geriatric risk scoring system developed in the USA was employed to validate mortality within 1-year of hospital discharge.


Introduction
Estimating mortality risk, is important in terms of medical decision-making and optimal management and care of older patients, especially those suffering from comorbid psychiatric illness (Kusumastuti et al., 2018).
Prognostic models for mortality of older adults have been developed for different populations (Levine et al., 2007). A search of a Scopus database for articles identi ed 103 articles describing 193 models. It was concluded that models were regularly developed to help with clinical decision making but their use is premature (Minne et al., 2011).
However, to the best of our knowledge, no speci c focus on the added burden of psychiatric comorbidity was tested (Ritt et al., 2017).
Psychogeriatric inpatients are more frail and complex to manage than geriatric patients due to higher rates of co-morbidity, polypharmacy and the adverse effects of life-long and current psychiatric morbidity (Soysal et al., 2017;Veronese et al., 2017). A prognostic index for 1-year mortality risk in older adults discharged from a general medicine service was developed based on administrative data (Levine et al., 2007). Risk factors independently associated with 1-year mortality included: age, length of inpatient stay, discharge to nursing home, congestive heart failure, peripheral vascular disease, renal disease, hematologic malignancy, and dementia. Administrative data validly identi ed high-risk mortality groups (Levine et al., 2007).
We aimed to validate a prognostic model for 1-year mortality in old-age psychiatry inpatient services.

Methods
This is a retrospective cohort audit in a general hospital inpatient psychogeriatric ward with unplanned psychiatric admissions during 2017.
We collected patients' gender, age, discharge location (home or facility), length of hospital stay and all risk factors tested by Levine et al (Levine et al., 2007). Data was collected from computerized discharge summaries.

Prognostic Risk Scoring System
The prognostic scoring assigned to each of the 9 nal risk factors are shown in table 1. A nal risk score was calculated by adding the points designated for each risk factor. For example, a 90-year-old patient (2 points) with congestive heart failure (1 point) and who is discharged to a nursing home (1 point) will have a nal risk score of 4 points.

Analysis.
Age was coded as per Levine's protocol (see table 1) as Zero (less than 70, One (70 to 74) and Two (75 or older). Items were collapsed to bivariate outcomes tested against survival using the chi-squared test. Any correlation from this tabulation with a p < 0.10 was entered into a Probit (probability unit) regression, and a de novo model established. The signi cance of the correlation in toto was tested using Wald's test and post hoc testing of the signi cance of each coe cient using in parallel, the risk scoring system for each participant was extracted using Levine's method (see table 1) and the correlation for this model was tested against survival.
Probit analysis operates like multiple regression with dependent or response variables that are binary. It enables converting data to a representation that could be viewed as a linear function. For both models, linear regression was used to estimate the proportion of variation attributable to the model (Larson et al., 2016).

Results
Between January 2017 and Dec 2017 there were 125 inpatient admissions to the closed psychogeriatric unit at the Dunedin Public Hospital. All 125 patients' electronic records were accessed and relevant information extracted. Mean age for the sample was 82.8 [+ 8.9] years, 82 (65.6%) women. Thirty-three (26.4%) [mean age, 87.7 (+ 11.1) years, 25 (75.7%) women] died before reaching the 1-year survival postdischarge threshold.
Using Levine's derivation cohort 1-year mortality risk scores (Levine et al., 2007), we demonstrated that a higher score was signi cantly predictive of mortality within 12 months of discharge (Wald c 2 = 9.1, df = 1, p= 0.0026). Patients who passed away had a mean score of 5.0 compared with patients alive after 12 months whose mean score was 3.3.
Two factors were statistically different between the "alive" and "passed away" groups: discharge to a care facility (Chi-square= 14.1; p<0.001) and the presence of metastatic cancer (Chi-square= 5.6; p<0.02).
There was a no correlation of length of stay over 5 days and mortality: in part because most patients stayed longer.
Three factors correlated signi cantly with mortality: discharge into care, a diagnosis of cancer, and metastatic cancer. These three factors were tested against each other using probit regression, and discharge to care remained highly correlated with mortality. The three factor model had a signi cant correlation using the Wald test (Wald c 2 =13.2, 3 df, p=0.0042)

Discussion
The last decade's studying high-dimensional data has driven an increase of prediction in medicine and psychiatry (Joyce & Geddes, 2020). We evaluated the ability of a published prognostic index for 1-year mortality of hospitalized older adults (Levine et al., 2007), using readily available standard administrative data, to predict mortality of inpatient psychogeriatric patients. The index was readily able to identify patients with a high 1-year mortality rate. Our ndings are also in line with other published studies that focused on psychogeriatric dementia inpatients after unplanned acute hospital admission (Sampson et al., 2013).
The present study has some major strengths and limitations. This is the rst study that has evaluated predictors of mortality in a cohort of hospitalized psychogeriatric inpatients. This was a single-centre study. The clinical pro le of the patients treated and cared for at psychogeriatric wards might be uneven among hospitals due to different policies dictating acute psychogeriatric services.

Conclusion
The Prognostic Risk Assessment Tool accurately predicted 1-year mortality in psychogeriatric inpatients. Further elucidation of the factors that predict mortality over time is important. A more hopeful compassionate approach to older people with an expectation of life measured in years not months may be a way forward.

Ethical Approval
Ethical approval was obtained from the University of Otago Ethics Committee reference number HD19/080 and the Southland Medical Foundation Ethics Committee project number 01603.

Ethics approval and informed consent
Ethical approval was obtained from the University of Otago Ethics Committee reference number HD19/080 and the Southland Medical Foundation Ethics Committee project number 01603. Consent for publication: Not applicable. Table   Table 1.