Integration of longitudinal deep-radiomics and clinical data improves the prediction of durable benefits to anti-PD-1/PD-L1 immunotherapy in advanced NSCLC patients

Table 4 Response prediction performance comparison between longitudinal and ensemble models in the independent test set for endpoint PFS6 and PFS9 by evaluating AUC, ACC, SENS, SPEC, PREC and bACC, respectively

Endpoint	Model	Features	N test	AUC [95% CI]	ACC [95% CI]	SENS [95% CI]	SPES [95% CI]	PREC [95% CI]	bACC [95% CI]
PFS6	Ensemble RF-baseline	DF-imm Clinical data	43	0.678 [0.513,0.836]	0.605 [0.442,0.744]	0.875 [0.731,1.000]	0.263 [0.071,0.467]	0.600 [0.436,0.758]	0.569 [0.448,0.684]
	Ensemble RF-longitudinal	DF-imm Clinical data	32	0.824 [0.658,0.953]	0.750 [0.594,0.906]	0.733 [0.500,0.938]	0.765 [0.533,0.947]	0.733 [0.471,0.933]	0.749 [0.594,0.897]
PFS9	Ensemble RF-baseline	DF-imm Clinical data	43	0.560 [0.377,0.731]	0.581 [0.442,0.721]	0.793 [0.643,0.933]	0.143 [0.000,0.364]	0.657 [0.487,0.811]	0.468 [0.360,0.590]
	Ensemble RF-longitudinal	DF-imm Clinical data	32	0.753 [0.549,0.931]	0.813 [0.656,0.938]	0.947 [0.826,1.000]	0.615 [0.357,0.889]	0.783 [0.609,0.950]	0.781 [0.631,0.923]

For each metric, the 95% confidence interval is shown and the highest value for each endpoint is highlighted in bold

ISSN: 1479-5876