Learning curve of laparoscopic duodenum-preserving total pancreatic head resection: a single-center dual CUSUM and risk-adjusted CUSUM analysis.

BACKGROUND

Laparoscopic duodenum-preserving pancreatic head resection (LDPPHR) is a parenchyma-sparing alternative to pancreaticoduodenectomy for benign and low-grade pancreatic head lesions. Although function-preserving, LDPPHR remains technically demanding, with clinically relevant postoperative pancreatic fistula (CR-POPF) and other complications. The learning trajectory and risk-adjusted quality profile of LDPPHR-t remain incompletely defined, particularly under evolving case-mix and postoperative care pathways.

METHODS

Consecutive adults undergoing LDPPHR-t were analyzed as a strictly time-ordered series. Efficiency learning was assessed using unadjusted CUSUM curves for operative time (OT) and estimated blood loss (EBL). Quality and safety learning were evaluated using risk-adjusted CUSUM (RA-CUSUM) for TO failure and CR-POPF (ISGPS grade B/C), with expected risks derived from multivariable logistic regression. Learning phases were delineated at concordant turning regions of OT- and EBL-CUSUM and interpreted alongside both RA-CUSUM curves. Sensitivity analysis used modified textbook outcome (mTO), defined as TO excluding 30-day readmission.

RESULTS

In 120 cases, Phase 1 (1-30), Phase 2 (31-80), and Phase 3 (81-120) were identified. OT decreased from 270 min (IQR 246.2-297.5) to 203 min (IQR 180.2-248.2) (P < 0.001), and EBL from 110 mL (IQR 100-200) to 50 mL (IQR 46.2-100) (P < 0.001). Case-mix demonstrated late-phase risk drift toward higher fistula-risk anatomy. Crude TO and mTO rates did not differ significantly across phases, but components diverged, with 30-day readmission increasing in the late phase while prolonged LOS declined to zero. In risk-adjusted monitoring, both TO failure and mTO failure RA-CUSUM showed a late-phase downward shift, whereas CR-POPF RA-CUSUM showed no sustained late upward accumulation.

CONCLUSIONS

A dual-track CUSUM and RA-CUSUM framework characterized LDPPHR-t learning under evolving anatomical complexity. Efficiency improved earlier, whereas risk-adjusted composite-endpoint stabilization emerged later. Concordant TO and mTO analyses support the robustness of the late-phase learning interpretation while clarifying the influence of post-discharge care pathways on crude composite-endpoint behavior.