Participant eligibility criteria for the asthma group were (1) age 13-17 years, (2) physician-diagnosed asthma for at least 1 year, and (3) ability to understand spoken and written English. The nonasthma group were age-matched adolescents with no current/past history of asthma and free of ongoing respiratory conditions. For both groups, we excluded those with other diagnoses producing respiratory symptoms (eg, upper respiratory infection, cardiac disease, cystic fibrosis) or significant cognitive impairment that could interfere with following the study protocol. Potential participants were recruited from the pediatric emergency department (ED) and outpatient clinics (primary practice and pediatric pulmonary practice) in a major university medical center located in the Northeastern United States. Of a total of 84 participants, most (73%, 61/84) were recruited from the ED and the remaining were from the study flyers (23%, 19/84) and clinician referrals from outpatient clinics (4%, 4/84). Unverifiable asthma diagnosis by medical records was the most common cause of ineligibility for the asthma group (n=52) followed by comorbidity with other respiratory diagnosis (n=14). For the nonasthma group, having an asthma diagnosis in the past (n=22) and presenting respiratory symptoms (n=7) were common reasons for ineligibility.

At enrollment, spirometry and FeNO tests were conducted for all participants followed by the measurement of asthma control and health care utilization for the asthma group. All participants used the device continuously for the next 7 days during which the asthma group completed daily the electronic asthma diary and the VAS in the morning and at bedtime. On completion of the 7-day trial, spirometry and FeNO tests were repeated for all participants. In addition, the asthma group completed a quality of life questionnaire and reported any health care utilization that occurred in the past week. Follow-up data on asthma control, health care utilization, and quality of life were collected at 3 months after the trial only from the asthma group. Only 2 of 42 participants in the asthma group were lost to the 3-month follow-up. Of the 40 follow-up cases, 30 were completed by mail and 10 were in-person with research staff. This study protocol was approved by the Institutional Review Board, the Research Subjects Review Board, located in the University of Rochester Medical Center, Rochester, NY, and informed consent and assent were obtained from parents and teens, respectively. The participants received a monetary incentive (US $130 for the asthma group; US $100 for the nonasthma group) for their participation.

All analyses were performed using SAS v9.3 (SAS Institute, Inc, Cary, NC, USA). Descriptive statistics were used to examine demographic and clinical characteristics. To assess concurrent validity, Pearson correlations were calculated between the device data from the asthma group with spirometer data (FEV₁ and FVC), FeNO, and other measures of asthma conditions, including asthma control questions, daily symptom diaries, VAS, quality of life, and health care utilization. To assess predictive validity of the device, regression was used to examine the relationships between the device data and asthma control, health care utilization, and quality of life collected at 3 months after the 7-day trial. Age, gender, and race were adjusted for in the regression analyses. Sensitivity and specificity were evaluated by assessing the device’s capability to classify 2 distinctive groups. A receiver operating characteristic (ROC) curve analysis was conducted to calculate the area under the curve (AUC) as an indicator of the device’s capacity to discriminate between asthma and nonasthma participants. An AUC of 1 indicates perfect classification and an AUC of 0.5 indicates that the ability to correctly classify is no better than chance. Based on the ROC, an optimum cutoff value was chosen to jointly maximize the sensitivity and specificity of the device. Given the few studies looking at devices for symptom monitoring in adolescents, any noticeable statistical trend is valuable; therefore, the significance level used in the analysis was less than .10. These findings are useful for an initial understanding of the validity of ADAM and for providing direction for further studies. Although it is understood that these findings will require further investigation, the risk of rejecting important research hypotheses was judged more important than the risk of type I error.

Details of participant flow from screening to the 3-month follow-up and sociodemographic characteristics of each group in the sample are reported elsewhere [34]. A total of 84 adolescents aged between 13 and 17 years (mean 15 years, SD 1.4) participated in the study. Of those, 61% (51/84) were females and 44% (37/84) were nonwhite adolescents. No significant differences were found between the asthma and nonasthma groups in age and gender. The asthma group included significantly more nonwhite adolescents, predominantly African American. At enrollment, the asthma group had slightly lower FEV₁ (mean 88.3%, SD 16.3 vs mean 90.9%, SD 13.8) and elevated FeNO (mean 28.6 ppb, SD 38.6 vs mean 25.6 ppb, SD 24.8) than the nonasthma group, yet these differences were not statistically significant. Within the asthma group, 19 of 42 (45%) reported active asthma symptoms at enrollment. Mean years since asthma diagnosis was 10.4 (SD 4.9) years. SABA use was reported by 95% (40/42) of the asthma group and most (60.5%, 25/42) were on at least one controller medication. The most common medication was inhaled corticosteroids (ICS), which was reported by 38% (16/42) of the asthma group, followed by ICS and long-term beta-agonist combination (29%, 12/42), and leukotriene modifier (21%, 9/42). No significant difference in control medication use was found between the symptomatic and nonsymptomatic groups. Only 2 participants (5%, 2/42) reported oral steroids as a current medication.

Descriptive analysis was conducted on data from the asthma group excluding 3 asthma teens for which no data were recorded in the device due to unknown mechanical issues. All 39 teens with asthma used the device for a mean 8.26 (SD 1.47) consecutive days (median 8, range 5-14 days) and each teen used the device for a mean 19.4 hours/day (SD 1.71; median 19.7, range 15-22 hours). When the number of coughs was compared for morning (6 am-11:59 am), afternoon/evening (noon-10 pm), and bedtime (10 pm-6 am), a greater number of coughs were registered during afternoon/evening compared to morning and bedtime (P=.01 and P=.004, respectively). The number of coughs was not significantly different by gender or age.

Greater number of coughs was correlated with poor asthma control and quality of life and higher use of health care services 3 months later (Table 2).

Table 3 presents the extent to which cough counts predicted asthma control, quality of life, and health care utilization after adjusting for age, gender, race, and family income. Coughs significantly predicted asthma control, and the regression model explained 42% of the variance in asthma control. Cough counts also significantly predicted the quality of life total score (β=-.55, P=.001) and each of the subscales including activity, symptoms, and emotional function 3 months later. Explained variance in quality of life was 38% for the quality of life total score, and 28%, 35%, and 41% for activity, symptoms, and emotional subscales, respectively. Health care utilization during the 3 months following the 7-day trial was significantly predicted by cough counts (β=.74, P=.004), explaining 76% of the variance. Particularly, coughs were associated with ED visits (r=.47. P=.004), asthma specialist visits (r=.45, P=.005), and office visits due to worsening asthma (r=.39, P=.02) that occurred in the 3-month period after the trial.

The capacity of the device to distinguish participants with asthma from those without asthma was assessed using ROC curve analysis. The ROC curve for the presence of asthma diagnosis had an AUC of 0.71 (95% CI 0.58-0.84). The AUC was significantly different from chance (χ² ₁=9.7, P=.002). The optimal cutoff value of the device was 0.56, with 51.3% sensitivity and 72.7% specificity (Figure 1). The cutoff value discriminating teens with asthma from those without asthma was translated into 0.83 coughs/hour or 19.92 coughs/day.

Table 4 summarizes findings pertaining to the validity of the device. Most of the expected relationships between ADAM data and conventional measures of asthma were substantiated by our findings.

Overall, we found positive temporal correlations, albeit low to moderate, between device data (ie, cough counts) and conventional measures of asthma symptoms and symptom control. Similarly, previous studies reported modest correlations between objective cough rates and subjective measures specific to cough (VAS for cough and cough scores) in individuals with asthma [49,50]. In our study, the generic self-report measures of asthma symptoms made it difficult to assess the degree of agreement between the device’s cough counts and the amount of coughs perceived by individuals. Nonetheless, when conceptualizing coughs as an indicator of asthma condition, use of generic measures of asthma control to establish the device’s validity as an asthma-monitoring tool can be justified. The demonstrated relationships between cough counts and concomitantly assessed activity limitation and other symptoms as well as overall asthma control provide support for the validity of the device as a tool for overall asthma monitoring rather than simply for cough. As in another study [49], the relationship between cough counts and quality of life suggest that experiencing coughing can take a toll on quality of life in teens with asthma. Moreover, the positive relationships between cough counts and health care utilization and school absenteeism during the trial could be further evidence that cough counts as measured by ADAM is a compelling indicator of asthma morbidity.

Unlike earlier reports of no relationship between spirometry data and coughs measured by an objective method [49,51] or self-report [52], we found that the cough counts by ADAM were associated with poor pulmonary function (as indicated by FEV₁ and FVC), suggesting that the number of coughs are indicative of airway obstruction. However, consistent with an earlier study [49], we were unable to establish the relationship between coughing and airway inflammation measured by FeNO. This may be because most (>50%) of the asthma group were asymptomatic (ie, no indication of active airway inflammation) before and during our trial period. Replication of the trial using a large number of patients with active symptoms is needed to determine the nature of relationship between coughs and airway inflammation.

This study demonstrated strong evidence for predictive validity of the device. After adjusting for sociodemographic factors, cough counts were predictive of asthma morbidity and quality of life at 3 months after the 7-day trial. The cough count predicted poor asthma control and quality of life at 3-month post-trial, accounting for 42% and 38% of the variance, respectively. Health care services used during 3 months after the trial were predicted by the cough count by our device and explained as much as 76% of the variance in acute health care utilization 3 months later. In an earlier study [53], worsening coughing was found to be predictive of severe asthma after 9 years in adult patients. Our findings support not only ADAM’s predictive validity, but also coughs as an important harbinger for upcoming deterioration of teens’ asthma morbidity that could undermine their overall well-being and impose serious burdens on the health care system. As such, the findings underscore the importance and need for a monitoring device such as ADAM that can increase patients’ awareness of the patterns of cough for early detection and has the potential for preventing serious and costly future consequences of asthma morbidity.

The device’s capacity to discriminate correctly the asthma group from those without asthma was assessed to determine sensitivity and specificity. We found poor diagnostic sensitivity of ADAM; that is, the chances of correctly identifying those with asthma were only 51% using 20 coughs/day as a cutoff. The poor diagnostic sensitivity of the device may have been due to the asymptomatic state of more than 50% of the asthma group in which the average cough counts were below the cutoff making it difficult for the device to appropriately identify asthma cases. In the future, studies maximizing differences between subsamples in symptom presentation will be essential for adequately assessing the sensitivity of the device. By contrast, ADAM demonstrated relatively better specificity such that the device correctly classified those who did not have asthma for 73% of cases. Using the device in an environment replete with everyday noises may result in a high number of cough counts exceeding the cutoff, even in those without asthma; thus, incorrectly excluding them from the nonasthma group. We observed occasional false events (2 coughs/hour) depending on environmental noises (70% sensitivity of the cough algorithm) [35], which is strikingly similar to 2.5 events/hour by the Leicester Cough Monitor [54]. Given the poor capacity to correctly classify patients with asthma, ADAM is not suitable for determining a clinical diagnosis of asthma, but is intended solely for monitoring cough in those with a confirmed asthma diagnosis. Like any monitoring device, false alarms are still an issue because these can cause unnecessary concerns or undermine users’ confidence in the monitoring tool. Therefore, further optimizing the accuracy of the monitoring device is warranted by refining the algorithm and adopting noninvasive techniques to minimize the influence of environmental noises (eg, direct application of an adhesive microphone on the chest wall).

Several limitations to this study’s design warrant caution. Because the device stored only the number of coughs without sound recording, validating the accuracy of cough counts through the manual confirmation of corresponding cough events was not done. Except for pulmonary function tests and FeNO, we primarily used self-report measures that were not specific to coughs to establish the validity of the device. The generic nature of the measure may have contributed to some of the nonsignificant correlations between the measures and device data. Although self-report measures are inherently subject to recollection bias, we attempted to address the challenge by strategically collecting daily symptom data electronically. In health care utilization data, we observed more than 90% agreement between self-report and medical record review. Therefore, it appears that recollection bias played little influence on our validity outcomes. Nonetheless, future research is needed to evaluate the validity of the device by simultaneously recording raw cough sounds and by using cough specific measures.

Moreover, distinction between the asthma and nonasthma groups was blurred because most of the asthma group did not present active symptoms during the trial, which may account for the device’s poor sensitivity. Also, use of the small and convenient sample of adolescents limited generalizability of the findings. In addition, the length of the 7-day trial might not be long enough to observe any meaningful changes in asthma symptoms, health care utilization, or biological measures including FEV₁ and FeNO. A longer observation period is warranted to assess the extent to which the device adequately captures changes over time in symptom patterns and other measures of asthma status. Lastly, there were a few technical challenges and limitations to the optimal operation of the device, which we discussed in our earlier report [34].

Several approaches have been attempted to develop technologies to monitor symptoms objectively, particularly coughs [45,49,50,54-58]. However, these existing approaches are considered unsuitable for ambulatory monitoring of symptoms due to practical challenges, including the laborious and time-intensive nature of processing audio data (so not fully automated and unable to provide real-time feedback to users), inability to monitor continuously beyond 24 hours, and the conspicuous appearance of the systems. For instance, the Leicester Cough Monitor [54,59,60] uses a similar sound recognition technical approach to that of ADAM. Unlike ADAM, in the Leicester Cough Monitor, the ambulatory component of the system consists strictly of an audio recorder and audio analysis is performed offline, which can take approximately 1 hour to process a 24-hour audio recording. This differs fundamentally from ADAM in which all of the processing/annotation of audio data are performed instantly on the mobile device and the results (cough counts) are provided as feedback in real time. Other cough detection systems [61] are not intended for ambulatory use or have yet to be validated in vivo for an extended observation period. In that sense, ADAM is the first ambulatory cough-monitoring device on a consumer mobile system with a capacity to fully automatize continuous real-time processing. This is also the first attempt to evaluate the validity of a cough detection device as a symptom-monitoring tool for several consecutive days in adolescents with asthma. Using a popular mobile device, such as iPod, as a platform for ADAM was well received by adolescents with asthma [34] and the majority of participants used the device daily for a week or longer period supporting ADAM as a sustainable asthma-monitoring tool for adolescents.

Overall, this study demonstrated the validity of ADAM as a symptom-monitoring device in teens with a confirmed asthma diagnosis. Poorly controlled asthma takes a toll on teens’ overall health and quality of life as well as on the health care system due to an increased economic burden associated with the use of urgent types of health care services. ADAM can potentially mitigate the adverse consequences by helping users detect and treat early symptoms before advancing to a poorly controlled state. The device is useful in increasing understanding about one’s current status of asthma control and in predicting asthma morbidity and quality of life for the near future. Such information can make the users become aware of symptoms and triggers and enable them to take appropriate and timely actions to address symptoms or prevent further deterioration of their symptoms. Objective symptom information from the device would be clinically useful in establishing optimum treatment plans and evaluating treatment effects. ADAM can be particularly useful and effective in monitoring symptoms occurring at night when environmental noises are minimal. Nighttime coughs were more common than wheezing [62] and worsening of asthma symptoms often occurs during nighttime [63]. As such, nighttime symptoms are often indicative of poorly controlled asthma [31,64], but patients often do not recognize or tend to neglect to report nighttime symptoms [65-68]. ADAM can be an invaluable tool that monitors nighttime symptoms, which would provide important clinical insight into the degree of symptom control and the response to asthma treatment.

Nonetheless, we caution that the device should not be considered as a diagnostic tool and its application to a broader age range remains to be evaluated. Although we suggest coughing as an important indicator of asthma morbidity, understanding the symptom in the context of other information, such as activity level and medication use, can potentially augment its clinical value and relevance. Originally, ADAM was designed to monitor symptoms and activity levels simultaneously capitalizing on the host system’s (iPod) built-in accelerometer to offer an insight into the nature of the relationships between symptoms and activity levels (eg, exercise-induced asthma). However, continuously recording and processing sounds and activity simultaneously quickly drained the platform battery, posing challenges for long-term uninterrupted sound processing. More study is needed to reassess the feasibility of concurrent operation of these 2 applications with optimized code and application power management. ADAM also allowed users to record the use of medications throughout the day to help them systemically review changes in symptom patterns in relation to medication use. This can motivate users’ adherence to treatment plans (when they noted symptom reduction after medication) or inform users or providers of the need for adjusting current medication or dosage (when no relief from symptom was achieved after medication). A clinical trial is needed to examine the effects of the device’s medication tracking function on users’ medication adherence.

ADAM represents an important undertaking in the field of mobile health (mHealth) that has exploded in the past decade to improve health outcomes and patients’ self-management capacity [69-71]. The application of mobile technologies in daily assessment of asthma symptoms can be a particularly attractive option for adolescents. Yet to be determined is the extent to which the appeal of mobile technology can translate into better symptom monitoring, awareness, and self-management behaviors, which ultimately leads to improved asthma outcomes.

The following sample clinical scenario demonstrates the potential application of ADAM in clinical practice: