In this retrospective observational study performed in a high-volume hepatobiliary-pancreatic unit, we evaluated the effect of a surgery-specific goal-directed therapy (GDT) physiologic algorithm on complications and length of hospital stay. We compared patients who underwent pancreaticoduodenectomy with either a standardised Enhanced Recovery After Surgery program (usual care group), or a standardised Enhanced Recovery After Surgery program in combination with a surgery-specific cardiac output-guided algorithm (GDT group). We included 145 consecutive patients: 47 in the GDT group and 98 in the usual care group. Multivariable associations between GDT and lengths of stay and complications were investigated using negative binomial regression. Postoperative complications were common and occurred at similar frequencies amongst the GDT and usual care groups: 64% versus 68% respectively, P=0.71; odds ratio 0.82; (95% confidence interval 0.39-1.70). There were fewer cardiorespiratory complications in the GDT group. Median (interquartile range) length of hospital stay was ten days (8.0-14.0) in the GDT group compared to 13 days (8.8-21.3) in the usual care group, P=0.01. Median (interquartile range) total intraoperative fluid was 3,000 ml (2,050-4,175) in the GDT group compared to 4,500 ml (3,275-5,325) in the usual care group, P <0.0001; but by day one, the median (interquartile range) fluid balance was similar (1,198 ml [700-1,729] in the GDT group versus 977 ml [419-2,044] in the usual care group, P=0.96). Use of vasoactive medications was higher in the GDT group. In our patients undergoing pancreaticoduodenectomy, GDT was associated with restrictive intraoperative fluid intervention, fewer cardiorespiratory complications and a shorter hospital length of stay compared to usual care. However, we could not exclude an influence of surgical caseload, which we have previously found to be an important variable. We also could not relate the increased hospital length of stay to cardiorespiratory complications in individual patients. Therefore, these observational retrospective findings would require confirmation in a prospective randomised study.