A Low-Cost, Cloud-Based Data Capture and Validation Pipeline for Postoperative Delirium Research

Abstract

Background: Prospective bedside data collection in perioperative research is labour-intensive and error-prone. Methods: We describe a workflow for prospective data capture in a 10-bed cardiac surgical intensive care unit: four Google Forms modules writing to Google Sheets, with Python/pandas scripts for processing and validation developed iteratively during data collection. Postoperative delirium (POD) served as an illustrative use case. Data quality issues were tallied from end-of-collection exports of the response sheets. Results: In 2021, the forms captured 7147 submissions for 762 patient identifiers; 610 enrolled patients formed the analytic cohort. The required-field completeness was 91.1–100% against the final form schemas and 99.7–100% relative to the form version in force at submission. Thirty-six submissions (0.5%) were affected by identical redundant entries, a non-numeric identifier, a postoperative-day value converted to a date by the spreadsheet, or manual flags. In the illustrative application, POD was recorded in 120 of 610 patients (19.7%). Conclusions: A free form-based front end with scripted validation can approximate several data quality functions of dedicated EDC platforms, but not their audit trails, role-based access or institutional governance. Its error detection performance was not evaluated against a reference standard.

The authors' abstract, as published at the source. BioMedInformatics, 2026 · DOI ↗