Evaluation of an ARIMA-Based Algorithm for Detecting Respiratory Events After Lung Transplantation: Case-Crossover Analysis with Descriptive Comparison to a Deviation-Based Model

Background: Home spirometry enables longitudinal monitoring of lung transplant recipients, but its utility is limited by measurement variability and reliance on fixed reference thresholds, which may fail to capture clinically relevant changes. We evaluated an autoregressive integrated moving average (ARIMA)-based algorithm for detecting respiratory events from daily home spirometry, alongside exploratory analyses of diagnosis-specific pre-event trends. Methods: We developed an ARIMA-based algorithm to identify abnormalities in daily FEV1 trajectories implemented within a web-based application. Performance was evaluated using a nested case-crossover design among 21 patients who experienced acute or chronic respiratory events. Case windows were defined as the 7 days preceding event onset, and control windows as 7-day periods occurring 90 days before the event. For each window, we assessed whether abnormalities were detected by the ARIMA-based algorithm and by a previously reported deviation-based algorithm. Associations between algorithm-detected abnormalities and respiratory events were estimated separately for two algorithms using conditional logistic regression. Additionally, we explored the number of detected events by specific clinical diagnoses. As an exploratory analysis, we examined the linear trends over three predefined pre-event intervals (0–14, 14–45, and 14–90 days before the event) for each individual to explore whether patterns differed by event type. Results: Of 21 events, both the ARIMA-based and deviation-based algorithms detected 11 events overall. Abnormalities identified by the ARIMA-based algorithm were strongly associated with respiratory events (OR 10.0; 95% CI 1.28–78.1; p = 0.028). The ARIMA-based algorithm identified most acute events (10/15; pneumonia and acute rejection) but detected few chronic or slowly progressive events (1/6; CLAD-related or pleural effusion). In contrast, the deviation-based algorithm detected fewer acute events (7/15) but captured more chronic or slowly progressive events (4/6). Exploratory trend analyses suggested that gradual FEV1 declines often preceded chronic events, indicating that trend features may aid discrimination diagnosis for events. Conclusions: The ARIMA-based algorithm may help detect acute respiratory events based on individual daily home spirometry FEV1 trend after lung transplantation. Incorporating trend features may complement detection of chronic or slowly progressive events. Combining short-term deviation detection with longer-term trend characterization may potentially support early event detection in home spirometry-based post-transplant surveillance. Clinical trial number : Not applicable.