Sleep monitoring challenges in patients with neurocognitive disorders: A cross‐sectional analysis of missing data from activity trackers

Abstract Background and Aims Activity monitors, such as Fitbits®, are being used increasingly for research purposes and data have been validated in healthy community‐dwelling older adults. Given the lack of research in older adults with neurocognitive disorders, we investigated the consistency of sleep data recorded from a wrist‐worn activity monitor in this population. Methods Fitbit® activity monitors were worn by hospitalized older adults as part of a parent study investigating sleep and step count in patients recovering from hip fracture surgery in a tertiary care academic hospital in Hamilton, Canada between March 2018 and June 2019. In this secondary analysis, we compared the proportion of missing sleep data between participants with and without a neurocognitive disorder and used a multivariable model to assess the association between neurocognitive disorder and missing sleep data. Results Of 67 participants included in the analysis, 22 had a neurocognitive disorder (median age: 86.5 years). Sleep data were missing for 47% of the neurocognitive disorder group and 23% of the non‐neurocognitive disorder group. The presence of a neurocognitive disorder was associated with an increased likelihood of missing sleep data using the Fitbit® activity monitors (adjusted odds ratio: 3.41; 95% confidence interval: 1.06–11.73, p = 0.04). Conclusion The inconsistent nature of sleep data tracking in hospitalized older adults with neurocognitive disorders highlights the challenges of using interventions in patient populations who are often excluded from validation studies. As opportunities expand for activity monitoring in persons with neurocognitive disorders, novel technologies not previously studied in this group should be used with caution.


| METHODS
The parent study, Researching the Effects of sleep on STep count dUring the Post-operative period (REST-UP), investigated the impact of sleep duration on outcomes in older adults recovering from hip fracture surgery. 8 We included 67 patients who were 65 years of age or older and admitted with hip fracture to the orthopedic ward of a tertiary academic hospital in Hamilton, Canada. Wrist activity monitors (Fitbit ® Alta 2) were applied to older adults, who provided informed consent themselves or via a substitute decision maker, after hip fracture surgery and worn for the duration of their hospital stay (or up to 14 days). We conducted an exploratory analysis of the device data from participants with neurocognitive disorders defined as: mild cognitive impairment (MCI/minor neurocognitive disorder), dementia (major neurocognitive disorder), Parkinson's disease and those who developed postoperative delirium. Preexisting diagnoses of MCI, dementia and Parkinson's disease were extracted from the chart from admission and consultation notes and postoperative delirium was extracted using the CHART-DEL (Chart-based Delirium Identification Instrument) method. 17 The average proportion of sleep data that failed to be recorded from participants with these conditions was compared to data from other participants in the study. We defined failure to record sleep data as the activity monitor not tracking sleep for one or more nights.
Differences in proportions were calculated using Chi-squared test and differences in means were calculated using the t test.
A multivariable model was created to evaluate the association between neurocognitive disorders and missing sleep data after adjusting for age, sex, and Charlson comorbidity index. All tests of significance were two-sided and p values less than 0.05 were consider significant. The statistical analysis was done using R   Table 1, including postoperative mobility and living environment. The proportion of missing sleep data was higher in those with neurocognitive disorders compared to those without neurocognitive disorders (47.0% vs. 23.0%, p = 0.004). There was a higher mean number of missing sleep days in those with neurocognitive disorders (3.32 vs. 1.49, p = 0.007), though the mean number of days participants wore the Fitbit ® was similar between both groups (6.86 vs. 6.89, p = 0.98). In the multivariable analysis (

| DISCUSSION
Study participants with neurocognitive disorders had more than three times the odds of having missing sleep data compared to participants without neurocognitive disorders, indicating that sleep may not be measured reliably by activity monitors in all populations. There are a number of possible explanations for this finding. First, participants with neurocognitive disorders were more likely to remove or dislodge the activity monitors at night. The removal of activity monitors by these patients has been previously reported 18 and presents a challenge to measuring sleep in this population, however, we cannot confirm whether this occurred in our participants as they were not constantly monitored at night. Second, in patients with neurocognitive disorders, the Fitbits ® may have had trouble establishing a "baseline period," which helps calibrate the device to measure sleep accurately. A recent systematic review investigating actigraphy in patients with Alzheimer's dementia, recommended a minimum 7-day period to establish an accurate baseline of sleep characteristics before collecting data. 19 Given the limitations of our inpatient study, we were unable to spend several days establishing a "baseline period" for patients. As the Fitbits ® were usually worn for less than 7 days due to short hospital stays, it is unlikely that baseline sleep characteristics were adequately established. Last, since wrist activity monitors track sleep based on heart rate and physical movement, 13 any increased movement during sleep could be associated with inaccurate recording of sleep. Abnormal movement during sleep is especially common in patients with neurocognitive disorders such as Parkinson's disease, 20 Lewy body dementia, 21 and hyperactive and mixed delirium. 22 As a result, increased movements at night in the neurocognitive disorder group could have contributed to the disproportionate missing sleep data between the two groups.  There are study limitations worth noting. First, we were unable to compare the data collected from the Fitbit ® Alta 2 with the goldstandard method of PSG. As a result, the reason for the disproportionate missing sleep data cannot be fully elucidated.
Second, the sample size of this analysis is small (n = 67), which is further divided between two groups, so our sample is unlikely to be representative of older adults hospitalized with neurocognitive disorders. The small sample size also precludes subgroup analysis of different neurocognitive disorders (i.e., dementia, delirium, etc.), which may have different patterns of missing sleep. Finally, while we presume that patients with delirium may have removed their activity monitors at night, however this could not be verified.
In conclusion, in this study of hospitalized older adults recovering from hip fracture surgery, sleep data measurement using wrist activity monitors were inconsistent in participants with neurocognitive disorders. This highlights the challenges of using devices that have been validated for use in healthy adults and applying them to older adults with multimorbidities. Alternate methods may be required to reliably and accurately evaluate sleep in future research investigations involving this patient population.