Intervention Targeting Beyond Risk Prediction
Knowledge database Technology Data management & digitalisation Human Workload & well-being C.2: Reduction of idle times of MRI systems in radiology departments through the use of AI-based scheduling software - MRIdleWhile prediction models can identify patients at high risk for no-shows, or more generally any adverse events, not all of these cases may be responsive to interventions. Targeting interventions solely based on predicted risk may limit real-world effectiveness. A dual-targeted approach, that considers both risk and ‘reachability’, could be beneficial to better use resources and increase potential impact.
Problem description, research question and relevance
Interventions such as reminder calls or follow-ups are often directed at those with the highest predicted risk. However, such outreach efforts may fail when some patients can’t be contacted, or may not engage with interventional efforts. This leads to wasted resources and limited returns, even when the predictive model is accurate. Without accounting for reachability, efforts may fall short.
Methods and procedures in the project
Instead of focusing solely on targeting patients or interest groups with the highest risk, intervention studies should also take into account which subjects or subgroups may be most likely to positively respond to such efforts. This could be based on:
- Whether the patient can be easily reached;
- Previous responsiveness of that patient to outreach;
- A predictive model which predicts the likelihood of a patients’ responsiveness, if no prior contact attempted;
- Simply asking the patient their contact method preferences (e.g. some patients don’t like to be called, and may respond better to an email or text message)
Results and findings
By combining the risk predictions from the prediction model with some data and information about the patients, as well as more ‘responsiveness’-specific data, targeted interventions could become more impactful. Mitigating the possibility of having issues in contacting patients would help reach a wider pool of patients for the same staff effort.
Recommendations for practice
- Incorporate information on reachability and prior engagement when designing interventions, rather than relying solely on predicted risk.
- Use past contact success, predicted responsiveness, and patient-stated communication preferences to guide who gets contacted and how.
- Combining risk scores with reachability data helps focus staff effort on patients who are both high-risk and likely to respond—maximizing impact without increasing workload.