Read thought provoking articles on respiratory healthcare management including COPD and Sleep Apnea.

Spotlight: Predicting Daily CPAP Compliance

[fa icon="calendar"] Aug 4, 2017 12:46:04 PM / by Uthara Keerthi

My name is Uthara Keerthi, and I’m a Data Scientist at Somnoware. Continuous positive airway pressure (CPAP) is one of the most commonly used methods to reduce the severity of sleep apnea. Despite its serious consequences on health, many patients fail to adhere to CPAP therapy on a regular basis. Several studies have been conducted to determine the important drivers for short-term and long-term therapy compliance.

At Somnoware, we studied the average daily CPAP compliance by the patient population for a 90-day period. Our motivation behind this was that any significant patterns in the overall compliance could give sleep physicians a better picture regarding the daily usage of the CPAP device by patients once they are put on therapy. This article focuses on the method, conclusions, and impact of our analysis.


We considered a sample of 14,682 patient records from the data collected from sleep centers and hospitals. The data contained patient details such as age, weight, AHI, trial compliance, titration AHI, and starting date of therapy. We also obtained information on the daily usage (in hours) of the CPAP device by each patient for the first 90 days of therapy. Since our goal was to find the average usage on each day, the records of those patients for whom usage data was unavailable for any day were discarded from the sample.

The day number (ranging from 1 to 90) and the daily CPAP usage were set as the independent and dependent variables for the analysis respectively. We then partitioned the data into training and test sets using a 30-70 split. The training set was used to build the model whose accuracy was then evaluated using the test set. The mean squared error (MSE) was used as the performance metric for the model. A low value of MSE indicates that the model under consideration is good at accurately predicting future results. We built 3 different linear regression models to determine the pattern in average usage and we chose the one with the lowest MSE. We observed an increase in usage for the first few days followed by a steady decrease over time. This aligned with our hypothesis that patients in general, fail to regularly adhere to therapy.


By setting this as the baseline model, we can apply further analyses for building predictive models as solutions to other problems. We can build a predictive model for CPAP therapy compliance for an individual patient by estimating the deviation in daily CPAP device usage from the baseline model. The daily compliance can be predicted more accurately by incorporating other relevant patient information such as age, AHI, trial compliance, and titration AHI into the model.

The individual patient compliance model can be further improved by including daily CPAP usage once therapy starts and data is available. Physicians can thereby, monitor their patients’ therapy adherence and take effective steps in case there is a likelihood that the usage drops after a certain period. These preventive measures to enhance therapy compliance will significantly reduce hospital readmissions, a major contributor to rising healthcare costs.

The model can also be used to predict whether a given patient will achieve Medicare compliance or not. This is estimated by predicting whether the patient will be compliant at least 70% of any 30-day window within the 3-month time under consideration. A patient who is Medicare compliant will have to pay only a fraction of the costs of renting the CPAP device and purchasing other supplies. In short, the application of predictive modeling for CPAP compliance can help in reducing healthcare costs in the long run.

 To learn how Somnoware helps facilities automate their testing processes, obtain predictive insights, and drive greater patient engagement, please select this link.

Topics: clinical sleep data, sleep center software, sleep center emr, sleep lab management, sleep medicine software, cpap adherence, sleep care management, sleep lab software

Uthara Keerthi

Written by Uthara Keerthi