![]() In our experiments, the doctors achieve an average diagnostic accuracy of 71.40%, while the associative algorithm achieves a similar accuracy of 72.52%, placing in the top 48% of doctors in our cohort. We compare the accuracy of our counterfactual algorithms to a state-of-the-art associative diagnostic algorithm and a cohort of 44 doctors, using a test set of 1671 clinical vignettes. To resolve this, we present a causal definition of diagnosis that is closer to the decision making of clinicians, and derive counterfactual diagnostic algorithms to validate this approach. We show that failure to disentangle correlation from causation places strong constraints on the accuracy of associative diagnostic algorithms, sometimes resulting in sub-optimal or dangerous diagnoses. Here, we argue that diagnosis is fundamentally a counterfactual inference task. Counterfactual inference sits at the top of this hierarchy, and allows one to ascribe causal explanations to data. As noted by Pearl, associative inference is the simplest in a hierarchy of possible inference schemes 24, 25, 26. This is in contrast to how doctors perform diagnosis, selecting the diseases which offer the best causal explanations for the patients symptoms. This raises the question, why do existing approaches struggle with differential diagnosis? All existing diagnostic algorithms, including Bayesian model-based and Deep Learning approaches, rely on associative inference-they identify diseases based on how correlated they are with a patients symptoms and medical history. Despite significant research efforts and renewed commercial interest, diagnostic algorithms have struggled to achieve the accuracy of doctors in differential diagnosis 17, 18, 19, 20, 21, 22, 23, where there are multiple possible causes of a patients symptoms. In particular, machine learning assisted diagnosis promises to revolutionise healthcare by leveraging abundant patient data to provide precise and personalised diagnoses 8, 9, 10, 11, 12, 13, 14, 15, 16. In recent years, artificial intelligence and machine learning have emerged as powerful tools for solving complex problems in diverse domains 5, 6, 7. These errors are particularly common when diagnosing patients with serious medical conditions, with an estimated 20% of these patients being misdiagnosed at the level of primary care 3 and one in three of these misdiagnoses resulting in serious patient harm 1, 4. In the US alone an estimated 5% of outpatients receive the wrong diagnosis every year 1, 2. Providing accurate and accessible diagnoses is a fundamental challenge for global healthcare systems. Our results show that causal reasoning is a vital missing ingredient for applying machine learning to medical diagnosis. ![]() ![]() ![]() While the associative algorithm achieves an accuracy placing in the top 48% of doctors in our cohort, our counterfactual algorithm places in the top 25% of doctors, achieving expert clinical accuracy. We compare our counterfactual algorithms to the standard associative algorithm and 44 doctors using a test set of clinical vignettes. To overcome this, we reformulate diagnosis as a counterfactual inference task and derive counterfactual diagnostic algorithms. We show that this inability to disentangle correlation from causation can result in sub-optimal or dangerous diagnoses. However, existing machine learning approaches to diagnosis are purely associative, identifying diseases that are strongly correlated with a patients symptoms. In medical diagnosis a doctor aims to explain a patient’s symptoms by determining the diseases causing them. Machine learning promises to revolutionize clinical decision making and diagnosis. ![]()
0 Comments
Leave a Reply. |
Details
AuthorWrite something about yourself. No need to be fancy, just an overview. ArchivesCategories |