Prepare II - Predictive Model Training and External Validation

Overview

The general aim of the project is training and testing a predictive model of inpatient rehabilitation stay after hip and knee replacement for osteoarthritis. This specific part focuses on data collection and analysis for the model validation.

The following variables will be investigated to evaluate their predictive value towards length of rehabilitation stay (primary outcome), place of discharge, burden of care at discharge and/or functional recovery (secondary outcomes): * Patient age at admission * Patient sex * Patient profession * Living status (e.g. home or residential unit) * Diagnosis * Surgical procedure * American Society of Anesthesiologists Score * Body Mass Index before surgery * Day of surgery * Duration of surgery (minutes) * Surgical access technique * Surgical ward stay * Pre-existing comorbidities * Pharmacological treatment of pre-existing conditions * Hemoglobin levels at inpatient rehabilitation admission * Pain (Numeric Rating Scale) at inpatient rehabilitation admission * Function - cognitive domains at inpatient rehab admission * Barthel Index of Activities of Daily Living (ADL) at inpatient rehab admission * Physiotherapic functional evaluations * Rehabilitation program * Function - cognitive domains at inpatient rehab discharge * Inpatient rehabilitation stay * Place of discharge * Use of walking aids at discharge * Need of assistance with the ADL at discharge * Barthel Index of Activities of Daily Living at inpatient rehab admission Due to retrospective study design, patients will not be recruited ex-novo and will be used data already stored in the internal databases. Data retrieved from clinical records and other sets internal to the hospital will be collected on a dedicated Excel sheet and divided per section of interest (baseline, intervention, outcomes). Data from 1710 patients who underwent total hip or knee replacement and subsequent inpatient rehabilitation in our hospital were collected, in the previous part of this project, to train the model. In this second part, the same variables from 400 patients will be collected to test the model predictivity. Put together, these samples represent around 80% and 20% of the total population (2100) respectively, which is a common cohort distribution to test for validity. To verify the validity (discrimination and calibration) of the newly developed prediction and stratification models, we will conduct a temporal external validation based on a sample of newly collected patients undergoing hip and knee at the same hospital but at a different point in time (2018, as compared to 2019 for the development sample). To this aim, we plan to include a convenience sample of at least 400 patients: the sample size was determined based on a minimum sample size calculation, using Hoeffding's inequality (DOI: 10.1016/j.jspi.2012.09.013), which is sufficient to identify differences in performance smaller than 10% performance points with confidence higher than 95% (minimum sample size=390). Basic modelling methodologies such as linear and logic regression will also be taken into consideration. The best performing model for each outcome of interest will be finally selected.

Conditions

Interventions

Eligibility

Locations (1)

Outcomes

Central Contacts