Importance: Refining eligibility guidelines may identify more appropriate patients to undergo useful medical procedures.
Objective: To improve cost-effectiveness in selecting patients with melanoma for sentinel lymph node biopsy (SLNB).
Design, setting, and participants: This hybrid prognostic study/decision analytical model was conducted among patients with melanoma who were eligible for SLNB at 2 melanoma centers from Australia and the US from 2000 to 2014. Participants consisted of 2 cohorts of patients with melanoma undergoing SLNB and a cohort of eligible patients without SLNB. Individualized probabilities of SLNB positivity generated by a patient-centered methodology (PCM) were compared with those generated by conventional multiple logistic regression analysis investigating 12 prognostic factors. Prognostic accuracy was assessed by the area under the receiver operating characteristic curve (AUROC) for each methodology and by matched-pair analyses.
Interventions: Triaging appropriate patients to undergo SLNB.
Main outcomes and measures: Total number of SLNBs performed (giving total cost) vs number of SLNB-positive outcomes (a measure of effectiveness) was evaluated. Improved cost-effectiveness through judicious patient selection was interpreted as increased numbers of SLNB-positive outcomes achieved, decreased numbers of SLNBs performed, or both outcomes simultaneously.
Results: Among 7331 patients with melanoma, SLNB outcomes were assessed in 3640 Australian patients (2212 males [60.8%]; 2447 aged >50 years [67.2%]) and 1342 US patients (774 males [57.7%]; 885 aged >50 years [66.0%]); 2349 patients eligible for SLNB who did not undergo the procedure were included in the simulation. PCM-generated probabilities achieved an AUROC of 0.803 in predicting SLNB positivity in the Australian cohort and 0.826 in the US cohort, higher than corresponding AUROCs generated by conventional logistic regression analysis. In simulation, adopting many SLNB-positive probabilities as minimally acceptable patient-selection criteria resulted in fewer procedures performed or increased the expected numbers of positive SLNBs. A minimally acceptable PCM-generated probability of 8.7% elicited the same number of SLNBs as historically performed (3640 SLNBs), with 1066 positive SLNBs (29.3%), constituting an improvement of 287 additional positive SLNBs compared with 779 actual positive SLNBs (36.8% improvement). In contrast, adopting a 23.7% PCM-generated minimum cutoff probability resulted in performing 1825 SLNBs, or 1815 fewer SLNBs than the actual experience (49.9%). It resulted in the same expected number of positive results (779 SLNBs), for a 42.7% positivity rate.
Conclusions and relevance: This prognostic study/decision analytical model found that the PCM approach outperformed conventional multiple logistic regression analysis in predicting which patients would have positive results on SLNB. These findings suggest that systematically producing and exploiting more accurate SLNB-positivity probabilities could improve the selection of patients with melanoma for SLNB compared with using established guidelines, thus improving the cost-effectiveness of the selection process. Eligibility guidelines to undergo SLNB should include a context-tailored minimum cutoff probability.