(A-162) Prospective Validation of the Adverse Drug Reactions in an Ageing Population Risk Prediction (ADAPTiP) tool and Comparison with Previously Described Risk Prediction Tools: Study Protocol

Background: Adverse drug reactions (ADRs) and adverse drug events (ADEs) are burdensome for health services worldwide. A number of ADR and ADE risk prediction models have been developed and validated but none are in use in regular clinical practice. The Adverse Drug Reactions and Events in an Ageing Population Risk Prediction (ADAPTiP) Tool was newly developed using logistic regression and 5-fold cross validation and resulted in area under the receiver-operator curve (AUC) of 0.75 in the ADAPT dataset (development dataset). External validation is essential to demonstrate reproducibility and reliability of a new clinical prediction tool and is a necessary step towards practical implementation.

Objectives: The primary aim of this study is to externally validate the newly developed ADAPTiP tool in a prospective, newly collated database with a secondary aim to compare it with previously published ADR and ADE risk prediction tools.

Methods: This will be a prospective cross-sectional study of acute medical admissions screened for a suspected ADR-related hospital admission using validated algorithms to assess the causality, preventability and severity of each reaction in addition to the causative medications. The target sample size is 2000 patients based on the calculation of 10 episodes per candidate variable, using the previously calculated ADR prevalence of 10% and the 20 candidate variables and will take place across two different hospital sites. All reporting will adhere to Transparent Reporting of a multivariable prediction model for individual prognosis or diagnosis (TRIPOD).

Results: Categorical data will be presented as frequencies with percentages and continuous data will be presented as medians with 25th–75th percentiles.
The cross validated ADAPTiP model will be applied in the newly collected dataset to measure the AUC. Other prediction models such as ADDROP, Geriatric ADE score and GerontoNET will also be applied in the dataset in order to compare the AUC between the models. Callibration plots will be made for all models. Performance will be defined as excellent when AUC = 0.90-1.00, very good when AUC = 0.80 to 0.89, good when AUC = 0.70-0.79, poor when AUC = 0.60-0.69 and unacceptable when < 0.60. Sensitivity and specificity were calculated for each of the tools.

Conclusions: We will present a large prospective validation study of a newly developed ADR risk prediction tool with comparison with previously published tools as per TRIPOD guidelines.