Systematic Undercounting of Steps by Smartphones Widespread
Smartphones and wearable devices used to measure physical activity may systematically undercount the steps of older people and women.
Smartphones and other wearable devices that measure physical activity may systematically undercount the steps of obese people, women, and people of various ethnic groups, according to the results of a study published in Medical Hypotheses.
Mobile devices are increasingly being used to track a person's physical activity. Embedded sensors and associated apps can provide enormous amounts of data for health research, but the accuracy of the findings generated by this big data has yet to be determined. Matthew A. Brodie, PhD, senior research officer at Neuroscience Research Australia in New South Wales, and colleagues sought to explore the extent to which this type of data source should replace more traditional, validated clinical assessments in health research. The effect that invalid conclusions can have on vulnerable populations is of particular concern.
Dr Brodie and colleagues hypothesized that "unvalidated physical activity smartphone apps may substantially and systematically underestimate activity in vulnerable population sub-groups across the world." The authors noted that considerable variations exist in the validity and test-retest reliability in recorded steps even for healthy young adults.
The investigators evaluated the step counting accuracy of a smartphone app used by 717,527 individuals from 111 countries, using data from 48 participants between the ages of 21 and 59 years with a body mass index (BMI) between 17.7 and 33.5 kg/m2. The researchers conducted 2 studies: the first in 2017 and the second in 2018. The 2017 study revealed a 15% to 66% undercounting by Apple iPhones, which was significantly poorer than the 38% to 100% accuracy for Android phones. The 2018 study showed a median accuracy of 38% to 105% for walks for iPhones, whereas the median accuracy for Android phones was 30% to 125%.
Although research has shown wearable devices to be generally accurate during stereotypical treadmill-like walking, both Android and Apple phones demonstrated large error ranges between 0% and 200% of the steps taken. The authors noted that the error ranges for the percentage of steps counted was higher for slower and shorter walks.
Among the study limitations were the small sample sizes and the problems associated with generalizing conclusions from groups of younger healthy subjects from middle-class backgrounds and European ethnicity to the overall population.
The authors stressed the importance of unbiased, accurate, and validated assessment tools to advancing scientific knowledge. They noted also that the risk exists that unconscious bias regarding typical behavior may be embedded into consumer-grade smartphone apps, favoring fast, treadmill-like walking. This may compromise the validity of measurements provided by these devices and result in low-quality big data.
Dr Brodie and colleagues concluded by advocating for more research to develop smartphone apps that work for diverse global populations.
Brodie MA, Pliner EM, Ho A, et al. Big data vs accurate data in health research: Large-scale physical activity monitoring, smartphones, wearable devices and risk of unconscious bias. Med Hypotheses. 2018;119:32-36.