Feasibility of text message sleep assessment in African American and Latino patients with type 2 diabetes

Abstract

Study Objectives:

Text messaging (TM) may provide an inexpensive and convenient method for self-reported sleep assessment. This pilot study evaluated the feasibility of a TM sleep diary among a racial/ethnic minority population with uncontrolled type 2 diabetes.

Methods:

A convenience sample of 40 participants with uncontrolled type 2 diabetes was recruited. Participants wore an Actiwatch (Philips Spectrum Plus, Philips Respironics, Murrysville, Pennsylvania) for 7 consecutive days during both wake and sleep intervals and completed a daily TM sleep diary including 10 questions adapted from the Consensus Sleep Diary. The relationships between sleep measures from TMs and actigraphy were explored through Bland-Altman plots and correlations.

Results:

Of the 40 participants enrolled, 34 were African American and 6 were Latino. The mean age was 52.2 years (standard deviation = 8.2), and the mean hemoglobin A1c was 9.0% (standard deviation = 1.5). All but 1 participant attempted to complete the TM sleep diary. With a maximum of 70 TM replies possible, the median number of responses per participant was 66 (interquartile range = 59.5–69). Actigraphy and TM measures were related for total sleep time (median = 382 vs 393 min, respectively [r = .71; P < .01]), sleep onset latency (median = 31.4 vs 27.5 min [r = .61; P < .01]), time in bed (433.3 vs 489.3 min [r = .74; P < .01]), and sleep efficiency (77% vs 86% [r = .45; P = .005]). The measure of wake after sleep onset was higher from actigraphy than from TM, with a weak relationship between the 2 measures (median 47.9 vs 6.0 min [r = .31; P = .05]).

Conclusions:

TM is a novel and feasible method for sleep assessment in racial/ethnic minority adults with uncontrolled type 2 diabetes.

Citation:

Biggers A, Henkins J, Barton I, et al. Feasibility of text message sleep assessment in African American and Latino patients with type 2 diabetes. J Clin Sleep Med. 2021;17(1):69–78.

BRIEF SUMMARY

Current Knowledge/Study Rationale: Since 2017, the American Diabetes Association has recommended sleep assessment in people with type 2 diabetes. In clinical practices with low resources where care is provided to racial/ethnic minorities with uncontrolled diabetes, newer, convenient, and inexpensive methods of sleep assessment are needed.

Study Impact: This novel study shows the feasibility of a text messaging sleep diary. However, challenges remain in the automated decoding of text message responses, and additional research is needed to understand the potential value of a text messaging sleep diary in practice implementation.

INTRODUCTION

Poor sleep is increasingly recognized as a risk factor for worsening glycemic control in adults with type 2 diabetes.^1–7 African American and Latino patients are more likely to report worse sleep quality and duration compared with non-Hispanic White patients.^8–10 Poor sleep health may contribute to worse hemoglobin A1c levels and higher rates of diabetes complications seen in African American and Latino patients.¹ Sleep assessments to screen for poor sleep are a necessary component in the overall care of these populations with type 2 diabetes based on recommendations from the American Diabetes Association.¹¹

However, sleep assessments are not widely performed. Only 43% of primary care physicians report asking patients about sleep compared to 80% who report asking about exercise and 79% who report asking about healthy diet.¹² Infrequent sleep assessment by primary care physicians occurs despite 84% of providers reporting an understanding of the relationship between sleep and chronic conditions.¹² One potential barrier to following the American Diabetes Association sleep guideline could be the challenge of obtaining convenient yet accurate sleep assessments.

The pencil-and-paper sleep diary is the gold standard for self-reported sleep assessments.¹³ However, paper sleep diaries have limitations. Primarily, their validity is challenged by recall bias. In addition, incompleteness, illegibility, or filling out the information all in 1 day, known as hoarding¹³ or parking lot syndrome,¹⁴ limit the usefulness and accuracy of paper sleep diaries. However, considering that they are inexpensive and can be widely used in diverse populations, paper diaries are quite useful when filled out correctly.

Text messaging (TM) for sleep assessments could be an inexpensive, convenient alternative to paper sleep diaries. Daily TM can provide real-time, time-stamped information on sleep and limit recall bias. TM can also eliminate issues with legibility and hoarding. In addition, TM is widely accessible, with 96%–98% cell phone ownership in minority populations.^15–17 TM sleep diaries have potential use in clinical sleep assessments, although they have not been studied. It is unclear how TM sleep diaries compare with more objective sleep measures (ie, wrist actigraphy).

The aim of this exploratory analysis is to evaluate the feasibility of TM sleep diaries. We compare self-reported (TM sleep diary responses) and objective (wrist actigraphy) sleep measures in African American and Latino patients with poorly controlled type 2 diabetes.

METHODS

Population and setting

This study was conducted in an academic medical center in Chicago, Illinois, with predominantly low-income, racial/ethnic minority patients. A total of 40 African American or Latino patients (aged 21–75 years) with poorly controlled type 2 diabetes (hemoglobin A1c > 8%) were recruited from a primary care setting. The participants were recruited from a concurrent randomized controlled trial on diabetes self-management using mobile health technology (ClinicalTrials.gov identifier: NCT02990299). In the randomized controlled trial, participants were randomized to receive standard diabetes care or enhanced diabetes care utilizing health coaches (community health workers) and telehealth pharmacist encounters. At routine data collection events, participants were invited to join the second, current study, and if they were interested and eligible then they were consented separately. Participants were also compensated $40 for completing this additional observational study.

Design

This study was an observational cohort study (n = 40). The study protocol was approved by the University of Illinois at Chicago Institutional Review Board (Protocol #2016-0380).

Inclusion criteria

To be eligible for the study, individuals had to meet the following criteria: (1) self-identified as African American or Latino/Hispanic, (2) verbal fluency in English, (3) aged 21–75 years, (4) history of type 2 diabetes for more than 1 year, (5) ownership and use of mobile phone with TM plan, (6) receipt of primary care within the University of Illinois Hospital & Health Sciences System for more than 1 year. Exclusion criteria included: (1) plans to move from the Chicago area within the next year, (2) living outside Chicago communities of recruitment, (3) living in a household with someone already in the study, (4) unable to read a TM on a mobile phone, and (5) unable to verbalize comprehension of the study or impaired decision-making. These criteria mirrored the criteria of the randomized controlled trial from which the participants were recruited for this observational study.

Procedures

Participants were asked to perform 2 procedures over a 7-day time period. First, they were provided with a wrist-worn actigraphy watch to wear for 7 days. Every day thereafter, participants were asked to respond to TMs regarding the previous night’s sleep.

Wrist actigraphy

Each participant received an actigraphy watch (Philips Spectrum Plus, Philips Respironics, Murrysville, Pennsylvania) worn on the nondominant wrist. The participants received a standardized tutorial on how to use the watch from the research personnel. They were instructed to press a watch button at night when going to bed and again in the morning upon awakening. At the end of the 7 days, participants returned the watch to research personnel for cleaning and data download of sleep measures via Philips Actiware software. Sleep measures in the Actiware software were manually coded based on (1) actigraphy watch button depression, and (2) activity counts and light intensity when the button was not pressed. The TM sleep diary (eg, bedtime and arise time) was not used to code actigraphy sleep parameters. Completion of 5 of the 7 days was required for calculating the actigraphy measures.

TM sleep diary

The TM sleep diary was adapted from the Consensus Sleep Diary.¹⁸ The 10 TM questions used in the modified TM sleep diary are noted in Table 1. Research personnel scheduled TMs using the custom software application mytapp. Upon enrollment, participants received a test TM to verify service and to confirm basic skills with using TMs. Participants received a paper-based handout of all the TM sleep diary questions to orient them to the subsequent TMs that they should expect. Research personnel reviewed the handout with the participants and answered any questions. The first TM was sent at approximately 10:00 am. This message was accompanied with instructions to use “am” or “pm,” “h” for hours, and “m” for minutes. Next, 9 questions were sequentially sent after responses were received from the participant. If participants did not respond to the diary by 2:00 pm, then they received a reminder telephone call.

Table 1.

TM sleep diary.

Question	Variable
1. What time did you get into bed?	Bedtime
2. What time did you try to go to sleep?	Sleep time
3. How long did it take you to fall asleep?	Duration to fall asleep
4. How many times did you wake up, not counting your final awakening?	Number of awakenings
5. In total, how long did these awakenings last?	Total awakenings
6. What time was your final awakening?	Wakeup time
7. What time did you get out of bed for the day?	Get out of bed
8. How would you rate the quality of your sleep? 1 = very poor, 2 = poor, 3 = fair, 4 = good, 5 = very good	Quality of sleep
9. How long did you nap or doze (during the day)?	Duration of nap
10. Comments (Anything unusual about last night’s sleep?)	Patient comments

TM = text message.

Data from the TM sleep diary were downloaded from mytapp to a .csv file for coding and analysis. A predesigned data format for TM responses was developed so that times and numeric values could be consistently analyzed. Two authors reviewed the TM data separately and coded and formatted the responses. A third author adjudicated any discrepancies. The following a priori rules were followed: (1) formatting was applied if absent in the response (eg, a response of “6pm” for bedtime was coded as “6:00 pm”); (2) if a question required a numeric value, then a response was coded with a common metric (eg, for the question “How long did you nap or doze (during the day)?” a response of “2 hours” was coded as “120 minutes”); (3) a response of “none” was coded as “0” minutes (eg, length of nap); (4) the average of 2 numeric values was used if a participant indicated a range (eg, “1–2 hours” was coded as “90 minutes”); (5) am or pm was corrected if necessary based on expected sleeping schedule; and (6) bedtime and sleep time responses were corrected when reversed, to ensure that bedtime preceded sleep time. The final TMs were converted to sleep measures.¹⁸ Responses for 3 of the 7 days were required for calculating the TM sleep diary measures.

Survey data collection

Baseline information included sociodemographic data, the risk for OSA assessed using the STOP-Bang questionnaire,¹⁹ and the use of pain, sleep, or psychotropic medications. Sleep quality information was obtained through the Pittsburgh Sleep Quality Index.²⁰

Statistical analysis

Calculated sleep measures included total sleep time (TST), sleep onset latency (SOL), time in bed (TIB), sleep efficiency, and wake after sleep onset (WASO). Descriptive statistics were calculated for demographic and sleep/wake measures. Pearson’s correlation coefficients were computed for sleep measures based on wrist actigraphy and the TM sleep diary. Bland-Altman plots were used to examine graphically whether there was any systematic bias of TM sleep diary when compared to actigraphy, to visualize the variability of the bias, to determine if the bias and variability were uniform throughout the range of measurements, and to identify outliers. R software²¹ (version 3.6.1, R Foundation for Statistical Computing, Vienna, Austria) was used for all analyses.

RESULTS

Study population characteristics

Table 2 displays the sociodemographic and health characteristics of the 40 participants in the study. Eighty percent of the participants were women. The mean (standard deviation) age was 52.2 (1.5) years, and the mean body mass index was 36.9 (8.5) kg/m². The average hemoglobin A1c was 9.0% (1.5). Eighty-five percent of the participants were African American, 15% were Hispanic/Latino, and 60% made less than $20,000 per year.

Table 2.

Study demographics and characteristics.

Variable	Overall (n = 40)
Age, mean y (SD)	52.2 (8.24)
BMI, mean kg/m2 (SD)	36.9 (8.47)
Hemoglobin A1c, mean (SD) (%)	8.9 (1.54)
PSQI score, mean (SD)	7.9 (3.60)
PHQ-9 score, mean (SD)	6.3 (5.91)
Sex, n (%)
Male	8 (20.0)
Female	32 (80.0)
Ethnicity, n (%)
Latino/Hispanic	6 (15.0)
African American	34 (85.0)
Education, n (%)
< High school	6 (15.0)
High school diploma/general education degree	14 (35.0)
Some college/graduated college	20 (50.0)
Income, annual, n (%)
> $10,000	16 (40.0)
$10,000–$20,000	8 (20.0)
> $20,000	14 (35.0)
Refused to answer	2 (5.0)
Risk for sleep apnea, n (%)
Low	18 (45.0)
Intermediate	1 (2.5)
High	21 (52.5)
Ever been diagnosed with OSA, n (%)
Yes	10 (25)
No	30 (75)
Sleep medication, n (%)
Yes	5 (12.5)
No	34 (85.0)
Do not know	1 (2.5)
Pain medication, n (%)
Yes	14 (35.0)
No	25 (62.5)
Do not know	1 (2.5)
Depression medication, n (%)
Yes	4 (10.3)
No	35 (89.7)

BMI = body mass index, PHQ-9 = Patient Health Questionnaire-9, PSQI = Pittsburgh Sleep Quality Index, SD = standard deviation.

The average (standard deviation) Pittsburgh Sleep Quality Index score among the participants was 7.9 (3.6; > 5 indicates poor sleep).²⁰ A total of 60% of the participants either reported a history of OSA or were at high risk based on the STOP-Bang screening.¹⁹ In addition, 12.5% of the participants took some form of medication to assist with sleep, 35% took medication for pain, and 10.3% took an antidepressant regularly. The mean (standard deviation) Patient Health Questionnaire-9 score was 6.3 (5.9; ≤ 4 indicates minimal depression, 5–9 indicates mild depression, 10–14 indicates moderate depression, and ≥ 15 indicates moderately severe to severe depression)²² among the participants.

Wrist actigraphy and TM sleep diary measures

Distributions of and comparisons between the sleep measures generated by the wrist actigraphy and TM sleep diary are shown in Table 3. There were correlations between actigraphy and TM in TST (median = 382 vs 393 min, respectively [r = .71; P < .01]), SOL (31.4 vs 27.5 min [r = .61; P < .01]), TIB (433.3 vs 489.3 min [r = .74; P < .01]), and sleep efficiency (77% vs 86% [r = .45; P = .005]). There was no significant correlation observed for WASO (median 47.9 vs 6.0 min [r = .31; P = .05]). Bland-Altman plots are shown in Figure 1, Figure 2, Figure 3, Figure 4, and Figure 5 to evaluate agreement among the TM sleep diary and wrist actigraphy sleep values. Comparisons between wrist actigraphy and TM sleep diary of the mean TST category are shown in Table 4. In addition, a mean SOL of < 8 minutes was found in 10% of patients by TM sleep diary vs 3% of patients by actigraphy.

Table 3.

Actigraphy and diary median, IQR, and correlation of sleep measures.

	Actigraphy Median IQR	Diary Median IQRa	Correlation 95% CI
TST (min)	381.8 (354.8–421.1)	392.8 (356.1–447.4)	0.71b (0.51–0.84)
WASO (min)	47.9 (35.7–68)	6.00 (2.9–17.6)	0.31 (–0.01 to 0.57)
TIB (min)	433.3 (402.6–485.5)	489.3 (436.1–535.4)	0.74b (0.55–0.85)
SOL (mins)	31.4 (21.8.7–51.6)	27.5 (15.1–44.5)	0.61b (0.30–0.78)
SE (%)	76.8 (72.5–82.3)	85.8 (72.5–89.1)	0.45b (0.15–0.67)

^aResponses for 3 of 7 days required for TM sleep diary measures. One patient had inadequate diary data needed for WASO, TIB, and SOL; 2 patients had inadequate diary data needed for TST and SE. ^bP values significant at < .05. CI = confidence interval, IQR, interquartile range, SE = sleep efficiency, SOL = sleep onset latency, TIB = time in bed, TST = total sleep time, WASO = wake after sleep onset.

Figure 1. Bland-Altman plot for total sleep time.

TM = text message.

Figure 2. Bland-Altman plot for sleep onset latency.

TM = text message.

Figure 3. Bland-Altman plot for sleep efficiency.

TM = text message.

Figure 4. Bland-Altman plot for time in bed.

TM = text message.

Figure 5. Bland-Altman plot for wake after sleep onset.

TM = text message.

Table 4.

Actigraphy vs TM sleep diary by category of TST

Mean TST	Actigraphy (n = 40)	TM Sleep Diary (n = 38)a
> 8 h	1	6
7–8 h	10	9
6–7 h	17	13
5–6 h	8	5
< 5 h	4	5

^aResponses for 3 of 7 days required for TM sleep diary measures. Two patients had inadequate diary data needed for TST. TM = text message, TST = total sleep time.

All 40 participants wore the wrist actigraphy for at least 5 days, and 39 participants attempted to complete the TM sleep diary (where there was at least 1 TM response). The total number of TM responses possible over the 7-day period was 70 (10 messages a day, including the optional comment at the end). The median number of TM responses was 66 (interquartile range = 59.5–69). Measures of agreement (correlations for continuous variables, Cohen’s kappa for discrete variables) comparing TM responses between the 2 independent coders ranged from 0.92–1 among 9 TM sleep diary questions (comments were excluded). Only 3 participants required a 1-time telephone prompt to complete the TM diary for a single occurrence. One person did not complete the TMs at all and was not prompted by research personnel because of scheduling obstacles.

DISCUSSION

In this observational study of 40 racial/ethnic minority participants with type 2 diabetes, the TM sleep diary was feasible and correlated with wrist actigraphy measures of TST, SOL, and TIB. The Bland-Altman plots showed general agreement in TST and SOL measures. However, the TM sleep diary overestimated TIB compared with actigraphy. The sleep efficiency and WASO measures were mildly related or unrelated between TM sleep diary and wrist actigraphy, with the TM sleep diary underestimating WASO compared with actigraphy.

To date, few studies have compared technology-based sleep diaries with paper sleep diaries and wrist actigraphy. One small observational study compared an electronic sleep diary application on an iPad to a paper sleep diary and actigraphy in 15 healthy participants using the Consensus Sleep Diary questions.¹⁴ Both the electronic and paper sleep diaries did not correlate with the actigraphy on various sleep measures (TST, SOL, TIB, sleep efficiency, and WASO). Nevertheless, the electronic and paper sleep diaries showed good correlation with each other. There were no significant differences between these 2 modalities for any of the sleep measures, supporting the value of technology-based sleep diaries as a sleep assessment tool.

In another small observational study, Jungquist et al²³ compared an electronic wrist sleep diary to both a paper sleep diary and wrist actigraphy. Thirty-five participants were instructed to log sleep information daily through a wrist-worn device and to fill out a paper sleep diary. Similar to our study, TST and SOL were moderately correlated between electronic sleep diary and actigraphy, and WASO was underestimated. Their study also supports technology-based sleep assessment.

The TM sleep diary could be a novel approach in collecting sleep measures. Our TM sleep assessment is user-friendly and is potentially a viable alternative to other technology-based sleep assessment tools, such as applications or wrist devices. In addition, cell phones are readily available and are widely used^16,17 by all racial/ethnic groups. Our study also shows high adherence rates using the TM sleep diary. In a previous observational study of 80 participants asked to fill out a pain diary, researchers found higher adherence rates among those using electronic diaries than among those using paper diaries.²⁴ More complete diary information increases its value.²⁵ Future studies tracking sleep among various populations are needed to validate the TM sleep diary and support its use.

Although the TM sleep diary shows promise, there remains limited research in minority populations. Sleep assessments are important to study in minority populations because of the higher chronic disease effect and health disparities in sleep in these populations. Furthermore, there are several studies^26–30 suggesting that self-reporting of sleep in racial/ethnic minority groups may not be as accurate compared to that of non-Hispanic White patients using current sleep assessment tools including the Consensus Sleep Diary. Several observational studies have shown stronger correlations between self-reported questionnaires^26–28 or sleep diaries²⁸ and wrist actigraphy^26–28 and/or polysomnography²⁸ among non-Hispanic White patients compared to African Americans. In addition, the Multi-Ethnic Study of Atherosclerosis and the Study of Latinos Sueño Ancillary Study of diverse populations showed that there are varying degrees of underestimation and overestimation of sleep duration when using self-reported questionnaires compared to using wrist actigraphy in African American and Latino cohorts.^29,30 These self-reported sleep measure discrepancies are problematic because epidemiological data on sleep differences among racial/ethnic groups often rely on self-reported data.³⁰ TM may provide a simple approach to sleep assessment in minority populations.

Our study has a number of strengths. We are the first to study the use of a simple TM sleep diary. All but 1 participant attempted to respond to the daily TM, with only 3 participants requiring a single phone prompt. Another strength of the study is that the timing of the TM sleep diary a few hours after participants’ final awakening, reduced the probability of recall bias and eliminated hoarding. In addition, the TM itself prompted the participants to answer the questions and resulted in higher diary compliance rates.^24,31 Furthermore, our study observed a racially/ethnically diverse population with a chronic health condition known to be adversely impacted by poor sleep. Additional studies in minority groups with TM solutions are needed.

Nevertheless, our study also has limitations. First, we cannot generalize our small sample size of urban African American and Latino patients with type 2 diabetes to the general population. Our participants likely represent highly motivated recruits from the concurrent randomized clinical trial, where they were required to own and use a cell phone (selection bias). In contrast, among lower-socioeconomic-level populations in general, there is a greater probability of cell phone service lapse and change in phone number.³² This trend would be unlikely to influence our study findings, given the eligibility requirements and the short duration of the study. However, lapses in cell phone service may reduce the adoption of the TM sleep diary in these populations.

In addition, wrist actigraphy may not accurately assess sleep as well as polysomnography; comparison to an objective gold standard assessment would be helpful. Another limitation of the study is that the TM sleep diary required coding. Most of the participants correctly answered the TM sleep diary questions (although in varying formats: eg, “8:00 AM”, “8 AM”, “8 in the morning”). Other responses required interpretation. Examples of challenges included TMs that (1) were unclear (eg, provided a written explanation instead of a numerical value), (2) had the wrong time of day in the answer (eg, switching AM with PM or vice versa), and (3) provided multiple answers that needed deciphering (eg, provided all the answers to the TM sleep diary questions under 1 response). To address this limitation, we double-coded the responses. Although sending and receiving TMs represents a low-cost approach to administering the TM sleep diary, response coding may increase the cost and resources required for adoption. Automated methods to interpret TM responses (eg, natural language processing) may be worth future exploration and study.

The results of this study could have clinical value. Using TMs is widely available, accessible, low-cost, and user friendly.^15,16 A TM sleep diary has the potential for use as a screening tool for racial/ethnic minority patients with diabetes and other chronic health conditions. The TM sleep diary may be used as the first step in identifying patients with sleep deprivation (< 7 hours) and poor sleep quality and may allow for further objective clinical sleep assessments and intervention. For future research, the TM sleep diary should be assessed in a larger sample size and provide different language options, such as Spanish, to evaluate its effectiveness more broadly.

DISCLOSURE STATEMENT

All authors have seen and approved the manuscript. Work for this study was performed at the University of Illinois at Chicago. This study was funded by a research supplement to promote diversity in health-related research grant from the National Institute of Diabetes and Digestive and Kidney Diseases (R01DK108141S) and a career development award from the National Heart, Lung, and Blood Institute (K01HL149775-01). The authors report no conflicts of interest.

ABBREVIATIONS

SOL

sleep onset latency

TIB

time in bed

TM

text message

TST

total sleep time

WASO

wake after sleep onset