Maximizing acceptance of clinical pharmacy recommendations to reduce length of hospital stay in a private hospital from Amman, Jordan

Abstract

Background: Clinical pharmacy interventions (CPI) usually require prior medical authorization. Physicians approve 80% of CPI and reject 20%. If pharmacists show that physicians should authorize all 100% CPI, the profession will step closer to a fully independent prescriber status. This study used an artificial neural network (ANN) model to determine whether clinical pharmacy (CP) may improve outcomes associated with rejected CPI.

Method: This is a non-interventional, retrospective analysis of documented CPI in a 100-bed, acute-care private hospital in Amman, Jordan. Study consisted of 542 patients, 574 admissions, and 1694 CPI. Team collected demographic and clinical data using a standardized tool. Input consisted of 54 variables with some taking merely repetitive values for each CPI in each patient whereas others varying with every CPI. Therefore, CPI was consolidated to one rejected and/or one accepted per patient per admission. Groups of accepted and rejected CPI were compared in terms of matched and unmatched variables. ANN were, subsequently, trained and internally as well as cross validated for outcomes of interest. Outcomes were length of hospital and intensive care stay after the index CPI (LOSTA & LOSICUA, respectively), readmissions, mortality, and cost of hospitalization. Best models were finally used to compare the two scenarios of approving 80% versus 100% of CPI. Variable impacts (VI) automatically generated by the ANN were compared to evaluate the effect of rejecting CPI. Main outcome measure was Lengths of hospital stay after the index CPI (LOSTA).

Results: ANN configurations converged within 18 s and 300 trials. All models showed a significant reduction in LOSTA with 100% versus 80% accepted CPI of about 0.4 days (2.6 ± 3.4, median (range) of 2 (0-28) versus 3.0 ± 3.8, 2 (0-30), P-value = 0.022). Average savings with acceptance of those rejected CPI was 55 JD (~ 78 US dollars) and could help hire about 1.3 extra clinical pharmacist full-time equivalents.

Conclusions: Maximizing acceptance of CPI reduced the length of hospital stay in this model. Practicing Clinical Pharmacists may qualify for further privileges including promotion to a fully independent prescriber status.

Keywords: Acceptance; Clinical pharmacy interventions; Hospitalization; Jordan; Length of stay; Neural networks; Patient readmission; Pharmacists; Private hospital.

Fig. 1

Summary of Inclusion of All Consecutive Patients with Documented Recommendation until final consolidation into one accepted and/or one rejected recommendation per patient per admission

Fig. 2

Artificial Neural Network Model Structure for the Length of Hospital Stay after the Index intervention. Endnote: ^*number of input nodes varied as we included the significant predictors or mismatched variables predicting the outcome

Fig. 3

Variable Impact for Length of Hospital Stay after the Index Intervention Artificial Neural Network, all 54 input variables included

Fig. 4

Variable Impact for Length of Hospital Stay after the index intervention, only mismatched 19 input variables included