Indirect Comparisons via Sorafenib for the Comparative Effectiveness of Two PD-1/PD-L1 Inhibitors to Treat Advanced Hepatocellular Carcinoma Patients without Prior Systemic Therapies

Figures & data

Figure 1 Flowchart of matching-adjusted indirect comparisons (MAIC), the validation of the MAIC method, and the second-order indirect comparison.

Table 1 Comparison of Characteristics Before and After Entropy Balancing

Characteristic	Sorafenib arm before EB (sample size = 533)	AB vs Sorafenib				LP vs Sorafenib
		AB arm (sample size = 336)^a	p-value of tests against the AB arm before EB	Sorafenib group after EB (effective sample size = 277)	p-value of tests against the AB arm after EB	LP arm (sample size = 100)^a	p-value of tests against the LP arm before EB	Sorafenib arm after EB (effective sample size = 65)	p-value of tests against the LP arm after EB
Age≥65 years (%)	36.21	48	<0.001	47.99	0.997	62	<0.001	61.70	0.957
Male (%)	84.24	82	0.157	82.01	0.998	81	0.041	80.97	0.997
Asian (%)	77.86	56	<0.001	56.03	0.993	28	<0.001	28.51	0.890
Black (%)	1.88	1.8	0.897	1.80	0.998	2	0.833	1.99	0.990
White (%)	19.51	37	<0.001	36.98	0.994	51	<0.001	50.65	0.955
ECOG PS=1 (%)	46.34	38	<0.001	38.01	0.997	38	<0.001	38.11	0.983
Extrahepatic spread (%)	64.73	63	0.405	63.00	0.999	52	<0.001	52.16	0.980
Macrovascular invasion (%)	27.95	38	<0.001	37.98	0.995	20	<0.001	20.07	0.983
Baseline AFP ≥ 400 ng/mL (%)	48.03	38	<0.001	38.02	0.994	30	<0.001	30.22	0.968
Etiology of hepatocellular carcinoma
Hepatitis B	54.03	49	0.020	49.01	0.998	19	<0.001	19.46	0.880
Hepatitis C	21.95	21	0.596	21.00	0.999	36	<0.001	35.80	0.978
Alcohol	14.26	10	0.005	10.01	0.992	28	<0.001	27.86	0.980

Note: ^aThe numbers were kept with the same decimal points as reported in original studies.

Abbreviations: AB, atezolizumab plus bevacizumab; LP, lenvatinib plus pembrolizumab; EB, entropy balancing; ECOG PS, Eastern Cooperative Oncology Group performance status; AFP, α-fetoprotein.

Figure 2 The kernel density of the entropy balancing weights of individual-level patients (IPD) in the sorafenib arm when balanced to the atezolizumab plus bevacizumab (AB) arm and the lenvatinib plus pembrolizumab (LP) arm. The deviation and dispersion of each line from the black vertical line demonstrate the unbalance of sorafenib IPD and the aggregate data of the corresponding comparator regimen before entropy balancing. The dissimilarity of the two lines of entropy balancing weights represents the unbalance between the aggregate data of the AB and LP arms.

Table 2 The Matching-Adjusted Indirect Comparison Estimates of Hazard Ratios of Overall Survival and Progression-Free Survival

Comparison	OS	PFS
AB vs sorafenib
MAIC, HR (95% CI)	0.58 (0.42–0.79)	0.59 (0.47–0.76)
IMbrave150 RCT, HR (95% CI)^a	0.59 (0.46–0.76)	0.59 (0.48–0.73)
p-value of comparing MAIC and RCT results	0.450	0.497
LP vs sorafenib
MAIC, HR (95% CI)	0.83 (0.52–1.32)	0.62 (0.41–0.94)

Notes: ^aThese estimates were from the IMbrave150 trial, which was an RCT study. The estimates were compared with the counterparts from the MAIC analyses, thereby validating or falsifying the specification of the MAIC analyses. The results suggested good validity of the MAIC analyses.

Abbreviations: OS, overall survival; PFS, progression-free survival; AB, atezolizumab plus bevacizumab; MAIC, matching-adjusted indirect comparison; HR, hazard ratio; RCT, randomized clinical trial; LP, lenvatinib plus pembrolizumab.

Table 3 Second-Order Indirect Comparisons of Atezolizumab Plus Bevacizumab vs Lenvatinib Plus Pembrolizumab

	OS	PFS
AB vs LP, HR (95% CI)	0.71 (0.42–1.23)	0.95 (0.60–1.51)

Abbreviations: OS, overall survival; PFS, progression-free survival; AB, atezolizumab plus bevacizumab; IPD, individual-level patient data; LP, lenvatinib plus pembrolizumab; HR, hazard ratio.

Table 4 The E-Values of Estimated Hazard Ratios in the Present Study

Comparison	OS	PFS
AB vs sorafenib	2.84 (1.85)	2.78 (1.96)
LP vs sorafenib	1.70 (1.00)^a	2.61 (1.32)^a
AB vs LP	2.17 (1.00)^a	1.29 (1.00)^a

Notes: Results are presented as E-value of HR point estimate (E-value of the upper bound of the 95% confidence interval). Per the definition of ^aBy definition, the E-value of the upper bound of a statistically insignificant effect estimate is 1.

Abbreviations: OS, overall survival; PFS, progression-free survival; AB, atezolizumab plus bevacizumab; HR, hazard ratio; LP, lenvatinib plus pembrolizumab.