Clinical Prediction Rules in Hospitalised Patients

Author: Kimberley Ruxton

Ruxton, Kimberley, 2018 Clinical Prediction Rules in Hospitalised Patients, Flinders University, College of Medicine and Public Health

This electronic version is made publicly available by Flinders University in accordance with its open access policy for student theses. Copyright in this thesis remains with the author. This thesis may incorporate third party material which has been used by the author pursuant to Fair Dealing exceptions. If you are the owner of any included third party copyright material and/or you believe that any material has been made available without permission of the copyright owner please contact with the details.


A significant number of clinical prediction rules (CPRs) have been developed for a wide range of medical conditions; however their routine clinical is limited due to lack of validation studies. Two CPRs were assessed for their predictive performance at Flinders Medical Centre (FMC): Wells and revised Geneva scores, used for assessing patients with suspected pulmonary embolism (PE), were assessed in an all-inclusive patient population; the Multidimensional Prognostic Index (MPI), based on a Comprehensive Geriatric Assessment (CGA), was assessed for predictive performance in a different geographical patient population.

Wells and revised Geneva scores were calculated in 1,724 patients referred to Flinders Emergency Department (ED) and FMC with suspected PE between January 2013 and May 2014. PE was confirmed using CTPA, V/Q scan, or compression ultrasound. Calibration and discrimination of the risk scores were evaluated. Subgroup analysis was conducted on patients assessed <24 vs. ≥24-hr from hospital presentation as well as patient hospital location (ED, medical, and surgical wards). The MPI score was calculated within the first three days in 737 patients admitted to FMC General Medicine or Acute Care of the Elderly (ACE) wards between September 2015 and February 2017. Discrimination of the MPI was evaluated for primary outcome, 6-month all-cause mortality, and secondary outcomes. Confirmatory and exploratory factor analysis (CFA and EFA) was conducted on the primary outcome for testing the dimensionality of the MPI. Additional analyses were conducted on three optimised versions of the MPI using the ARS score or RUDAS score or in combination.

PE results: Observed and predicted PE prevalence within each risk category (low, intermediate, and high) was similar for all three categories in the Wells and revised Geneva scores. The area under the ROC curve was 0.61 for the Wells score and 0.62 for the revised Geneva score. These results are substantially lower than in the derivation studies for the Wells and revised Geneva scores. Area under the ROC curve for patients assessed after 24 hours (Wells, AUC 0.56; revised Geneva, AUC 0.59) was substantially lower than patient assessed within 24 hours (Wells, AUC 0.62; revised Geneva, AUC 0.64).

MPI results: The MPI as either continuous or categorical variable was associated with 6-month mortality (MPI continuous: OR 2.34; MPI categorical, Mild: reference group; moderate: OR 2.97; severe: OR 5.06). The area under the ROC curve for 6-month mortality was 0.63. These results are substantially lower than in the derivation study. Optimised versions of the MPI did not differ to the original MPI. CFA showed poor model fit in a one-dimensional MPI model. EFA identified a two-factor model: one factor related to physical function and the other comorbidities. The goodness of fit tests indicated good model fit for the two-factor solution.

This study highlights several concerns regarding the routine use of original CPRs in different geographical patient populations or in an all-inclusive patient population which the derivation studies did not assess. It also highlights the importance of validating CPRs in large prospective multicentre studies with populations representative of those for which the tools will be used.

Keywords: Clinical prediction rules, Pulmonary embolism, Comprehensive geriatric assessment

Subject: Clinical Pharmacology thesis

Thesis type: Doctor of Philosophy
Completed: 2018
School: College of Medicine and Public Health
Supervisor: Professor Arduino Mangoni