This study focuses on the use of a new analysis method, called regression discontinuity or “RD” in the medical specialty for cancer diseases. Our project aims to help provide more evidence on the use of a standard blood test called prostate-specific antigen (PSA) to detect prostate cancer at an early stage. RD aims to determine whether patients benefit from this diagnostic test. While clinical trials (that is, a highly controlled study in which patients are randomly assigned to receive a treatment or not) are considered to be the method of choice to test for effective treatment, they may not fully capture real-life effects occurring during routine care. In addition, many clinical trials are limited in their time frame and, therefore, cannot be used to evaluate treatments in the long-term. RD achieves its aim by exploiting thresholds in variables that influence whether someone receives a treatment or not (e.g., tumour size). By “zooming in” around the threshold, it can be assumed that people just above and just below the threshold are similar to each other. Consequently, they can be compared in the same way as patients in a treated or untreated group of a clinical trial. We aim to investigate whether the method can potentially be applied for prostate cancer. Thus far, RD has not been applied to electronic health record data in the field of cancer care.
Regression discontinuity (RD) design is a quasi-experimental method that takes advantage of decision rules that assigns patients to a clinical intervention if they fall above/below a cut-off point. RD can be used to assess the causal treatment effects in clinical medicine. This study seeks to determine the applicability and feasibility of RD in electronic health record (EHR) data in the field of clinical oncology. Specifically, we aim to (1) determine whether PSA thresholds are associated with a change in the probability of receiving a clinical intervention (here prostate biopsies and referral to urologists), (2) evaluate if patient characteristics are balanced within a small bandwidth surrounding these thresholds, and (3) investigate whether associations between the clinical intervention and patient outcomes are robust to different choices of bandwidth around the threshold. Exposure variables include tumour types, laboratory and physical measurements (e.g., body-mass index, age, lactate dehydrogenase, and haemoglobin). Patient outcomes include oncology-specific measurements and outcomes such as number/duration of hospitalisations, (tumour) pain, overall survival/mortality (overall and by cause of prostate cancer). We will estimate “fuzzy” RD models using local linear regression to avoid overfitting data and triangular weights to give more influence to observations close to the threshold. In addition, we will use a mean squared error (MSE) optimal bandwidth that is empirically derived. We assess the sensitivity of the results using alternative bandwidths (e.g. bandwidths that are 50%, 75%, 125%, and 150% of the empirically derived mean squared error-optimal bandwidth). If feasible and applicable, RD analyses in EHR could generate valuable insights into the real-life effects of clinical intervention and help identify heterogenous treatment effects across more granular patient subpopulations, particularly in individuals that are normally excluded from clinical trials.
Primary outcomes will include mortality from CPRD. We are not planning to use ONS data as previous research has shown a strong overlap between CPRD and ONS mortality data.[30] We provide further detail on this below.
We aim to include secondary outcomes that can be measured in primary care, including referral to a specialists (CPRD), the diagnosis of prostate cancer (CPRD), inclusion into active surveillance programs (CPRD). We will also include subgroups of hospitalization relating to complications from prostatectomy using HES data.
Till Bärnighausen - Chief Investigator - University of Heidelberg
Maximilian Schuessler - Corresponding Applicant - University of Heidelberg
Anant Jani - Collaborator - University of Oxford
Christian Bommer - Collaborator - University of Heidelberg
Julia Lemp - Collaborator - University of Heidelberg
Justine Davies - Collaborator - University of Birmingham
Min Xie - Collaborator - University Hospital Heidelberg
Pascal Geldsetzer - Collaborator - University of Heidelberg
Sebastian Vollmer - Collaborator - Georg-August-Universität Göttingen
HES Admitted Patient Care;Patient Level Index of Multiple Deprivation;Practice Level Rural-Urban Classification