Evidence-based approach for the generation of a multivariate logistic regression model that predicts instrument failure

Objective: Identification of instrument failure (IF) represents a point to improve the quality of services provided by medical laboratories. Here, a logistic regression model was created to define the relationship between instrument downtime and laboratory quality management systems. Methods: Interval-level quality control (QC) and categorical quality assurance data from 3 identical chemistry analyzers was utilized to generate a logistic regression model able to predict IF. A case-control approach and the forward stepwise likelihood-ratio method was used to develop the logistic regression model. The model was tested using a case-control dataset and again using the complete sample. Results: A total of 650 downtime events were identified. A total of 22,880 QC data points, 187 calibrations, 24 proficiency testing events, and 107 maintenance records were analyzed. The regression model was able to correctly predict 59.2% of no instrument downtime events and 69.2% of instrument downtime events using the case-control data. Using the entire data set, the sensitivity of the model was 69.2% and the specificity was 58.2%. Conclusion: A logistic regression model can predict instrument downtime nearly 70% of the time. This study acts as a proof of concept using a limited data set collected by the chemistry laboratory.