Investigating the feasibility of crisis-discharge decision-support to reduce readmission rates at a psychiatric ward.

[摘要] ENGLISH ABSTRACT: The pressure on the availability of beds in South African psychiatric hospitals is high. Inresponse, the Western Cape has implemented a \crisis discharge policy. This policy is differentfrom the planned short stay practice as the patients are discharged earlier than what is clinicallyideal to allow people on the waiting list to be admitted. The crisis-discharge policy thereforestrives to optimise the combined healthcare outcome for the patient population (including thosecurrently in the ward and those on the waiting list for admission). Discharge and admissiondecisions are informed by reviewing clinical indicators of the patient population.A previous study conducted at Stikland Psychiatric Hospital, which is also the institution wherethis research was undertaken, reported crisis-discharge to be a significant predictor of an increasedrisk for readmission. This suggests that implementing a crisis-discharge policy to alleviatethe pressure on available beds, may in fact exacerbate the scenario. Currently, unaideddecision making is implemented by the clinical psychiatrists to solve this combinatorial optimisationproblem and it is therefore unlikely that the daily decisions are optimal.This study investigates readmission at Stikland Psychiatric Hospital, specifically at the acutemale inpatient ward to (i) determine whether variables exist that indicate that certain patientswithin this population have a higher risk of requiring readmission after a crisis-discharge; and,if such variables do exist, (ii) to determine the predictive capability of these variables with aview of recommending the feasibility of a decision-support system for crisis-discharge at themale inpatient ward. Various patient variables such as age, diagnosis, place of follow-up andsubstance use are analysed.Basic descriptive methods, biostatistics and data mining were employed to analyse the data.Predictive models were developed and evaluated using, amongst others, classi cation and regressiontrees and random forests. The research was conducted with continuous input fromclinical subject matter experts.The most important statistically significant variables pertaining to the risk of readmission arethe diagnosis, whether a patient belongs to a community after-care programme, and the areathat a patient originates from. Direct admissions and schizophrenic patients were found to betwice as likely to be readmitted as patients who are not from these groups. The schizo-affectiveand bipolar diagnostic groups are about three times as likely to be readmitted compared topatients who are not from these diagnostic groups. The substance induced psychosis diagnosticvariable, and a community programme variable indicated that patients were less than half aslikely to require readmission. These results are some of the insights that are presented in thisresearch project.The best-performing predictive model is able to classify whether patients would require readmissionfollowing a crisis-discharge with average accuracy of 70%. Based on these findings, furtherresearch towards to the development of a crisis-discharge decision-support tool is recommended.

[发布机构] Stellenbosch University