Similarly, companies asking gig workers to send pictures with masks, share location data and body temperature during Covid-19 also puts them at risk of loss of privacy, the research report, prepared by Aapti Institute and commissioned by the UNDP under the Business and Human Rights in Asia programme funded by the European Union, has pointed out.
While AI was being used in several hospitals for many purposes – from reading scans to predicting risks — Covid-19 has scaled up its use, with hospitals even detecting the extent of lung damage.
The report, prepared by researchers Aishani Rai, Vinay Narayan and Sarayu Natarajan, based on talking to experts, stakeholders and research said lack of data availability arises due to structural issues of digital inaccessibility.
“For instance, only 14% of Indian adult women owned smartphones in comparison to 37% of adult men. Data in the healthcare sector is also generated from health apps on smartphones that constantly monitor consumer behaviour. As smartphone data comes primarily from men with above-average incomes, overreliance on this data may distort our understanding of the health needs of women and of poor women in particular.”
When data about certain populations does not exist in sufficient quantities, it leads to “uninformative predictions for minority populations”, leaving predictions applicable to majority populations. The report cites how patients of lower socioeconomic status receive fewer diagnostic tests and medication due to under-represented datasets.
“Each country has its own patterns of diseases commonly prevalent. In India, cardiovascular diseases affect people much earlier than in middle and high income countries. Given that doctors usually diagnose heart attacks based on symptoms experienced by men, any AI developed to diagnose heart attacks will under-diagnose Indian women. Diverse datasets must not only capture diverse populations but also capture other socio-economic modalities that affect health. Prediction based solely on EHRs fails to capture external factors such as access to housing and transportation facilities that affect health,” it said, pointing out that healthcare facilities in India are concentrated in urban areas and the rural masses are bound to travel long distances by subpar means of transportation.
The research recognised that predictive analytics has immense potential to transform healthcare services by reducing rural patient deaths, citing nearly 4,300 people die every day due to poor diagnosis in India.