Despite early interest and clinical potential, a recently published literature review suggests that additional research into adherence and engagement is still needed to confirm whether diagnoses of mood disorders using smartphone voice recording strategies have a place in healthcare.

In the study, Harvard T.H. Chan School of Public Health and Beth Israel Deaconess Medical Center researchers parsed nine bibliographic databases and 1,845 abstract results to find scientific articles with relevant empirical data. The concept, they wrote, builds upon the relationship between speaking speed, volume, and pitch and various disorders, which has already been established using recordings from vocal exercises and therapy sessions.

“In the past few years, a plethora of smartphone applications for health have emerged, yet little evidence has been published in peer-reviewed journals regarding their scientific rigor,” the researchers wrote in the Psychiatric Rehabilitation Journal. “In contrast, the popular media has focused more on the promises of smartphone audio data to diagnose various diseases.”

The final review included seven studies, all of which provided participants with a study phone, as opposed to a personal phone, to collect data. The studies employed various conversational scenarios to collect the audio samples, and did not examine speech content when determining outcomes.

The researchers found that although smartphone applications are capable of collecting these data, it is still unclear whether the strategy would work in practice. Studies included in the review were largely exploratory, they wrote, and as a result employed small sample sizes. Behaviors could change when asking patients — in particular those with bipolar disorder — to use their personal phones instead of those provided for the study, and some patients were found to switch off their voice recording sensors to preserve their battery charge. Furthermore, data provided variable clinical utility between collection settings, with certain studies providing unexpected or weak findings.

All of this is not to say that smartphone monitoring is unfeasible — in fact, the researchers noted data suggesting that a phone’s GPS, accelerometer, and other sensors could provide “superior performance” when identifying and predicting clinical outcomes. However, more robust data and real-world evidence will be needed before smartphone audio recording can move into care.
“Many fundamental scientific questions still need to be addressed before [smartphone audio collection] can become a validated clinical tool,” they wrote. “While smartphone-based voice technologies hold great potential for mental health … practitioners should assume caution if considering using them today as part of illness management.”

Last year, Boston-based Sonde Health licensed voice-analyzing technology from MIT’s Lincoln Laboratories with the goal of passively monitoring phone users’ auditory biomarkers in realtime. San Francisco-based startup Ginger.io raised $20 million in 2015 to continue its focus on passive smartphone data collection for both mental and physical health.

Use of voice analysis to monitor non-mental health has seen some promising findings, including in the diagnosis of heart disease, as described in a 2016 Mayo Clinic study, and in Parkinson’s disease, where researchers demonstrated a 99 percent rate of accuracy back in 2013.