LAUSR

Intradialytic hypoxemia is a common but often overlooked condition in patients undergoing maintenance hemodialysis (HD), potentially contributing to adverse clinical outcomes. Sleep apnea, a condition marked by intermittent "saw-tooth patterns" (STP) in arterial oxygen saturation (SaO2), may underlie some cases of intradialytic hypoxemia. Identifying these patterns during HD could provide critical insights into the prevalence of intradialytic sleep apnea and its clinical implications. This study aimed to estimate the distribution of SaO2 STPs based on the start time of HD treatments. We analyzed intradialytic SaO2 data from a large cohort of HD patients across the United States. SaO2 was continuously monitored during HD sessions using the Crit-Line Monitor (CLM; Fresenius Medical Care North America), an FDA-approved device for measuring oxygen saturation and hematocrit in the extracorporeal dialysis circuit. SaO2 data were recorded at 10-second intervals and transmitted to a secure cloud-based platform for analysis. Data was collected between January 14, 2021, and July 30, 2023. SaO2 levels were tracked continuously during the HD sessions. To construct the one-dimensional convolutional neural network (1D CNN), a random sample of HD sessions was chosen. The time series data from these sessions were segmented into consecutive 5-minute intervals, and each segment was independently reviewed by a physician and a researcher who categorized them as either containing a saw-tooth pattern or not. The trained 1D CNN was then applied to the entire dataset to classify each segment as exhibiting an STP or not. Figure 1 provides a visual example of SaO2 measurements during a single HD session, highlighting segments classified as STP by the 1D CNN. The timing of these segments was categorized based on the hour during which the HD treatment began. For example, all 5-minute segments from treatments starting between 6:00 and 6:59 were assigned to the 6 hours. The STP rate was then calculated for each hourly category. Following rigorous data cleaning, the dataset included more than 3 million HD sessions from over 49,000 patients treated at 714 clinics. Analysis revealed that 11% of the 5-minute segments were classified as STP, while 89% were non-STP. The proportion of 5-minute segments exhibiting STP exceeded 15% in treatments starting between 4:00 and 4:59 local time. Conversely, the lowest STP rates were observed in treatments beginning between 14:00 and 15:59. These findings were further detailed in Fig. 2A, which shows the total number of 5-minute segments analyzed alongside the number of STPs for each hourly category, and Fig. 2B, which illustrates the hourly STP rates. Night, defined as a start time between 20:00 and 5:00, was associated with a higher prevalence of STP at 13.36%, compared to 10.77% during the daytime. Patients who initiated HD at night exhibited a significantly greater occurrence of STP (P < 0.00001). Our results indicate a high rate of STP in patients who have started HD during the nighttime. Since patients may be more likely to sleep during night and since STP of SaO2 are a hallmark of sleep apnea, we hypothesize that sleep apnea is a likely condition in these patients. As a next step, patients with a high rate of SaO2 STP should undergo formal testing for sleep apnea and, if confirmed, receive adequate treatment.