Written By: Advata Research Team
Healthcare is a deeply complex system where stakeholders - patients, clinics, doctors, payers, families, and governments - interact in ways that are hard to predict1. Add to this the varied nature of illnesses and healthcare delivery, and it soon becomes clear that healthcare problems require powerful solutions. Today this means integrating both artificial intelligence (AI) and advanced analytics to discover the insights needed for comprehensive applications2,3.
We believe that we have multiple responsibilities when using AI and advanced analytics to make decisions. This belief has given rise to Advata’s organization-wide initiative, “Responsible AI (RAI)”.
We use frameworks, practices, and tools that help us to responsibly design, develop, deploy, and monitor AI and advanced analytics systems throughout our work. And we use AI to maximize positive impact while ensuring ethical outcomes.
A Safer AI For Healthcare
Studies of AI in healthcare have shown how AI can help improve processes, enhance shared decision-making, and improve patient outcomes.
More recently, we learned that some earlier studies did not consider the impacts of hidden, unintended consequences or inequities5.
Led by concerns about unreported biases and the applicability of published predictive models, numerous frameworks have been developed to assess the risk of bias in healthcare data4. However, these have often been an after-the-fact assessment of published models. In addition, risks arise not only with specific AI algorithms but also across various stages of the data collection, aggregation, transformation, and consumption stages6.
The current state of AI in healthcare motivated our RAI initiative to establish our vision: At every stage of the customer journey, from problem definition, through phases of platform and solution creation, to customer deployment, we aim to:
- Define AI risks.
- Measure risks in the solution processes and analytics outcomes.
- Mitigate these risks and harms in deployments through collaboration, monitoring, and shared knowledge.
To help achieve Advata’s RAI vision, we defined six RAI pillars:
- Explainability: refers to AI models being explainable of data and outcomes, i.e., understanding what factors drive the model's predictions.
- Fairness: is described in terms of differences in model performance across various attributes like race, ethnicity, gender, etc.
- Robustness: refers to model performance in case of a lack of sufficient data.
- Transparency: refers to whether the AI system is transparent to the relevant stakeholders.
- Privacy: corresponds to aspects of data, insights, and model output that should not be shared with certain entities.
- Security: refers to where the data, the models, and the infrastructure is monitored and secured against unauthorized access and abuse.
Advata’s RAI journey has just started, and we invite people across the AI, healthcare, technology, legal, financial, and social fields to share your RAI journey with us. Together, we can responsibly create safer, more effective, and more equitable healthcare systems.
1. Institute of Medicine (US) Committee on Quality of Health Care in America. Crossing the Quality Chasm: A New Health System for the 21st Century. National Academies Press (US); 2001. Accessed June 5, 2021. http://www.ncbi.nlm.nih.gov/books/NBK222274/
2. “To Err Is Human: Building a Safer Health System” at NAP.Edu. doi:10.17226/9728
3. Bodenheimer T, Sinsky C. From Triple to Quadruple Aim: Care of the Patient Requires Care of the Provider. The Annals of Family Medicine. 2014;12(6):573-576. doi:10.1370/afm.1713
4. Moons KGM, Wolff RF, Riley RD, et al. PROBAST: A Tool to Assess Risk of Bias and Applicability of Prediction Model Studies: Explanation and Elaboration. Ann Intern Med. 2019;170(1):W1. doi:10.7326/M18-1377
5. Allen CA, Kumar V, Roche C, et al. An Analysis of Fairness and Bias in Predictions of Shoulder Arthroplasty Outcomes. Published online May 21, 2021. Accessed May 21, 2021. https://openreview.net/forum?id=-l2ZljKGfTr
6. Ahmad MA, Eckert C, Allen C, Kumar V, Hu J, Teredesai A. Fairness in Healthcare AI. In: 2021 IEEE 9th International Conference on Healthcare Informatics (ICHI). IEEE; 2021:554-555. doi:10.1109/ICHI52183.2021.00104