Implausible algorithm output in UK liver transplantation allocation scheme: importance of transparency

• Published in: The Lancet (British edition) Vol. 401; no. 10380; pp. 911 - 912
• Main Authors: Attia, Antony, Rowe, Ian A, Harrison, Ewen M, Gordon-Walker, Tim, Stutchfield, Ben M
• Format: Journal Article
• Language: English
• Published: England Elsevier Ltd 18.03.2023; Elsevier Limited
• Subjects: Algorithms; COVID-19; Decision analysis; Decision making; Hepatocellular carcinoma; Humans; Liver cancer; Liver diseases; Liver Transplantation; Resource allocation; Simulation; Tissue and Organ Procurement; Tissue Donors; Transplantation; Transplants & implants; United Kingdom; United Kingdom; United Kingdom > UK; United States > US
• ISSN: 0140-6736; 1474-547X; 1474-547X
• DOI: 10.1016/S0140-6736(23)00114-9

Algorithm-based allocation of resource-limited health-care interventions is growing; however, concerns over transparency and bias have restricted its use.1 Transparent algorithms can be readily explained, allowing patients and clinicians to clearly understand the basis for decision making.2 In 2018, the Transplant Benefit Score (TBS) was introduced to allocate deceased donor livers to patients with chronic liver disease and primary liver cancer (hepatocellular carcinoma) on a national basis. The TBS algorithm uses seven donor and 21 recipient parameters to predict the difference in survival without transplantation (need) to that after transplantation (utility) for each potential recipient (TBS=utility–need).3 Balancing the risk to benefit ratio between patients with chronic liver disease and patients with cancer, which typically arises on a background of chronic liver disease, is challenging.4 National reports show that for the first 3 years of the TBS scheme (excluding the period when TBS offering was suspended due to COVID-19), patients with cancer were rarely allocated livers by the TBS model and that waiting list removals for death or deterioration were considerably increased compared with patients with chronic liver disease alone (relative risk=1·58, 95% CI 1·22–2·06; appendix p 1).5 We aimed to understand TBS-derived allocation decisions with deterministic simulation methods. Taking these simulated patients with chronic liver disease alone and adding cancer counterintuitively reduced the probability of an organ offer being made, due to the TBS prediction that cancer improves survival without transplantation (relative cancer effect [IQR]: small cancer=2·08 [1·38–5·05]; large cancer=1·49 [1·00–3·78]; multiple cancers=2·07 [1·38–5·01]; figure A–C).