Table A2:

GRADE Evidence Profile for Independent Prognostic Value of MRD for Relapse

Number of Studiesa (Design)	Risk of Bias	Inconsistency	Indirectness	Imprecision	Publication Bias	Upgrade Considerations	Quality
Relapse (End-induction)
6 (observational)	Serious limitations (–1)b	No serious limitationsb	No serious limitations	Serious limitations (–1)c	Undetected	None	⊕⊕ Low
Relapse (End-consolidation)
2 (observational)	Serious limitations (–1)d	No serious limitations	No serious limitations	No serious limitationse	Undetected	None	⊕⊕⊕ Moderate
Relapse (After re-induction)
1 (observational)	No serious limitations	No serious limitations	No serious limitations	Serious limitations (–1)f	Undetected	None	⊕⊕⊕Moderate
Relapse (Pre-HSCT)
2 (observational)	No serious limitations	Serious limitations (–1)g	No serious limitations	No serious limitationsh	Undetected	None	⊕⊕⊕ Moderate

Abbreviations: GRADE, Grading of Recommendations Assessment, Development, and Evaluation; HSCT, hematopoietic stem cell transplant; MRD, minimal residual disease.

^aA body of longitudinal cohort studies is considered to begin as high-quality evidence for prognosis according to GRADE guidance for assessment of evidence about prognosis detailed in Iorio et al.⁵³

^bAll studies were prospective cohorts; median follow-up time was adequate in most studies; five studies^{56,58,66,67,70} did not measure or adjust the analysis for all key confounders. See Table A1 for further detail on risk of bias assessment.

^cResults across studies varied because of methodological heterogeneity, especially in thresholds for MRD positivity and adjusting for confounders precluding appropriate pooling. Relapse rate was low (range 12.0%–32.3%) and could decrease the statistical power, especially in smaller studies.^66,67 However, all effect estimates were in the same direction with overlapping confidence intervals (CIs) and were clinically significant.

^dBoth studies were prospective cohorts; median follow-up time was adequate but did not measure and/or adjust the analysis for all key confounders. See Table A1 for further detail on risk of bias assessment.

^eNo pooled estimate could be obtained because of methodological heterogeneity, and relapse rate per study was low (range 21.3%–25%), yielding questionable power. Thus confidence in the effect estimates is uncertain. However, clinical significance of the upper and lower bound of all CIs is retained despite wide CIs.

^fAdequacy of follow-up is unknown, as the median was not reported. Sample size was very small (n = 35) with a relapse rate of 43%; thus, CIs are wide and there is uncertainty in the point estimate, though the clinical decision would be the same and clinical significance is retained at the upper and lower bound.

^gPoint estimates across studies varied by approximately 3-fold, and analyses were adjusted adequately for similar confounders; however, both effect estimates were in the same direction and were statistically and clinically significant, with overlapping CIs.

^hConfidence intervals were very wide for both point estimates; adequate power to detect a clinically meaningful difference is unlikely because of low relapse rate (range 25.0%–26.8%), yielding uncertainty However, the clinical decision would be the same and clinical significance is retained at the upper and lower bound.