Comparative effectiveness of single-stage versus two-stage revision for hip prosthetic joint infection: emulation of the INFORM target trial using electronic health record data

OBJECTIVE: To estimate the comparative effectiveness of single-stage versus two-stage revision for hip prosthetic joint infection (PJI) in Chinese tertiary hospitals by emulating the INFORM randomized trial using the target trial emulation framework. DESIGN: Retrospective cohort study emulating a target trial, reported per the TARGET guideline. The protocol and analysis plan were written before analysis but were not prospectively registered. SETTING: Three tertiary hospitals in China. Index revision surgeries were performed between 1 January 2018 and 30 June 2022, with follow-up data locked on 31 December 2023, ensuring a minimum potential follow-up of 18 months for every patient. PARTICIPANTS: 754 adults with confirmed hip PJI undergoing revision surgery (280 single-stage; 474 two-stage), identified from electronic health records and confirmed by manual chart review. Treatment groups reflected the intended surgical strategy documented before or at the index operation. MAIN OUTCOME MEASURES: Primary outcome: Harris Hip Score (HHS) at 18 months. SECONDARY OUTCOMES: surgically treated reinfection, reoperation, mortality, and complications. The primary analysis used inverse probability of treatment weighting (IPTW) on the full cohort with multiple imputation of missing covariate and outcome data; 1:1 propensity score (PS) matching (280 pairs) was confirmatory. Patients who died before an assessment were excluded from the functional analysis at that assessment rather than having a score imputed, so the functional estimand is restricted to survivors (727 of 754 at 18 months). No a priori sample size calculation was performed because all eligible patients were included; precision is reported in place of retrospective power. RESULTS: Mean age was 60.6 (SD 14.4) years; 511 of 754 patients (67.8%) were male. Avascular necrosis was the leading indication for the primary arthroplasty (390/754, 51.7%). The adjusted mean difference in HHS at 18 months was 1.68 points (95% confidence interval [CI] - 0.12 to 3.48; P = 0.07), favoring single-stage but not reaching statistical significance; the interval is small relative to the within-person minimal clinically important improvement (MCII) for the HHS reported after primary hip arthroplasty (approximately 16 to 18 points), but that threshold describes improvement within an individual patient rather than a difference between groups, and no minimal important between-group difference has been established for this instrument in revision surgery, so we do not claim to have excluded a clinically important difference. At 3 months, single-stage was associated with a higher HHS (mean difference 5.82, 95% CI 1.43 to 10.21; P = 0.01); this early difference is attributable in part to many two-stage patients not yet having undergone reimplantation. Surgically treated reinfection was similar in frequency (13.2% v 11.8%; risk ratio [RR] 1.12, 95% CI 0.72 to 1.74). Fewer intraoperative adverse events were recorded for single-stage (RR 0.46, 95% CI 0.26 to 0.82), but this comparison is confounded by unequal operative exposure, because two-stage patients underwent two operative episodes and single-stage patients one, and it should not be read as a safety advantage of single-stage surgery. Mortality did not differ significantly (2.9% v 4.0%). The direction of the estimated difference was consistent across sensitivity analyses and no analysis was statistically significant, but the magnitude was not stable (1.18 to 2.48 points). Findings were directionally consistent with the INFORM trial, although HHS and WOMAC are distinct instruments and the estimates are not numerically comparable. CONCLUSIONS: In this observational emulation, an intended single-stage strategy was associated with 18-month functional outcomes that did not differ significantly from an intended two-stage strategy. The comparison is statistically inconclusive for small differences and does not demonstrate equivalence, which was not formally tested. Because treatment was not randomized, the protocol was not prospectively registered, the magnitude of the estimate varied across model specifications, outcome ascertainment was incomplete, and the functional comparison is conditional on survival, these findings are hypothesis-generating and require confirmation in prospective, adequately powered studies before informing practice.