Bayesian methods: a potential path forward for sepsis trials

Tomlinson, George; Al-Khafaji, Ali; Conrad, Steven A.; Factora, Faith N. F.; Foster, Debra M.; Galphin, Claude; Gunnerson, Kyle J.; Khan, Sobia; Kohli-Seth, Roopa; McCarthy, Paul; Meena, Nikhil K.; Pearl, Ronald G.; Rachoin, Jean-Sebastien; Rains, Ronald; Seneff, Michael; Tidswell, Mark; Walker, Paul M.; Kellum, John A.

doi:10.1186/s13054-023-04717-x

Critical Care

Table 1 Examples of posterior results for potential Tigris outcomes analyzed with different weights on the prior, and with observed PMX absolute risk reductions of 0–20%

From: Bayesian methods: a potential path forward for sepsis trials

Observed mortality in 100 PMX versus 50 control	Use of historical data in prior	Adjusted by APACHE		Unadjusted
Observed mortality in 100 PMX versus 50 control	Use of historical data in prior	OR	Pr(OR < 1)	OR	Pr(OR < 1)
24% versus 44% (ARR 20%)	100% fixed weight	0.46 [0.28, 0.76]	100.0	0.53 [0.33, 0.85]	99.5
	75% fixed weight	0.45 [0.27, 0.76]	99.9	0.51 [0.31, 0.84]	99.6
	Commensurate	0.44 [0.21, 0.85]	99.3	0.47 [0.23, 0.84]	99.5
	None (Tigris only)	0.38 [0.16, 0.88]	98.9 (a)	0.40 [0.19, 0.82]	99.3 (a)
29% versus 44% (ARR 15%)	100% fixed weight	0.52 [0.32, 0.84]	99.6	0.59 [0.36, 0.93]	98.8
	75% fixed weight	0.52 [0.31, 0.87]	99.4	0.58 [0.35, 0.96]	98.2
	Commensurate	0.52 [0.28, 0.95]	98.3	0.56 [0.39, 0.98]	97.9
	None (Tigris only)	0.50 [0.24, 1.04]	96.9 (b)	0.52 [0.26, 1.05]	96.6 (b)
33% versus 44% (ARR 11%)	100% fixed weight	0.56 [0.35, 0.92]	99.0	0.64 [0.40, 1.03]	96.8
	75% fixed weight	0.56 [0.34, 0.95]	98.4 (c)	0.63 [0.39, 1.05]	96.2
	Commensurate	0.57 [0.31, 1.07]	96.1	0.64 [0.35, 1.13]	94.0
	None (Tigris only)	0.59 [0.28, 1.22]	92.3 (d)	0.62 [0.31, 1.26]	90.9
37% versus 44% (ARR 7%)	100% fixed weight	0.63 [0.39, 1.00]	97.6	0.69 [0.43, 1.08]	94.7
	75% fixed weight	0.64 [0.38, 1.07]	95.7 (e)	0.70 [0.42, 1.15]	92.1
	Commensurate	0.66 [0.36, 1.28]	89.7	0.71 [0.40, 1.29]	87.5
	None (Tigris only)	0.73 [0.35, 1.56]	79.9 (f)	0.75 [0.37, 1.49]	79.9
39% versus 44% (ARR 5%)	100% fixed weight	0.65 [0.40, 1.05]	96.2	0.72 [0.45, 1.13]	92.1
	75% fixed weight	0.67 [0.40, 1.12]	93.8	0.73 [0.44, 1.20]	89.5
	Commensurate	0.69 [0.39, 1.31]	88.1	0.76 [0.42, 1.39]	83.0
	None (Tigris only)	0.76 [0.37, 1.60]	76.6	0.81 [0.42, 1.63]	72.3
41% versus 44% (ARR 3%)	100% fixed weight	0.66 [0.40, 1.09]	94.8	0.74 [0.47, 1.18]	89.5
	75% fixed weight	0.69 [0.40, 1.17]	91.4	0.76 [0.46, 1.27]	85.5
	Commensurate	0.72 [0.40, 1.41]	83.8	0.80 [0.45, 1.47]	77.2
	None (Tigris only)	0.83 [0.40, 1.72]	69.3	0.88 [0.45, 1.75]	63.5
43% versus 44% (ARR 1%)	100% fixed weight	0.70 [0.43, 1.12]	93.1	0.77 [0.49, 1.22]	86.6
	75% fixed weight	0.73 [0.43, 1.22]	88.7 (g)	0.79 [0.49, 1.29]	82.1
	Commensurate	0.80 [0.45, 1.55]	75.5	0.85 [0.48, 1.60]	70.9
	None (Tigris only)	0.93 [0.46, 1.88]	57.6	0.95 [0.48, 1.92]	55.6
44% versus 44% (ARR 0%)	100% fixed weight	0.72 [0.44, 1.16]	91.0	0.79 [0.50, 1.23]	85.0
	75% fixed weight	0.74 [0.44, 1.26]	86.5 (h)	0.81 [0.49, 1.32]	79.8
	Commensurate	0.85 [0.46, 1.73]	68.3 (i)	0.88 [0.50, 1.68]	66.0
	None (Tigris only)	1.02 [0.49, 2.15]	47.8	1.00 [0.50, 2.00]	49.8

Examples of analyses of potential Tigris results showing observed absolute risk reductions of 0% to 20% from an observed control group event rate of 44% (22/50). Both adjusted and unadjusted analyses are shown, for fixed weights of 100%, 75%, and 0% (analysis of Tigris only) on the prior from the treatable cohort and the commensurate prior. The results show the odds ratio (OR) and 95% CrI, along with the posterior probability that the OR is < 1. Each shaded block of results is based on the same simulated dataset; all data sets share a single set of APACHE II values, with identical sample means and standard deviations in the PMX and control groups. For cell annotations (a–i), see the results section

Back to article page

ISSN: 1364-8535

Contact us

Submission enquiries: journalsubmissions@springernature.com