Background: Disinvestment (removal, reduction, or reallocation) of routinely provided health services can be difficult when there is little published evidence examining whether the services are effective or not. Evidence is required to understand if removing these services produces outcomes that are inferior to keeping such services in place. However, organisational imperatives, such as budget cuts, may force healthcare providers to disinvest from these services before the required evidence becomes available. There are presently no experimental studies examining the effectiveness of allied health services (e.g., physical therapy, occupational therapy, and social work) provided on weekends across acute medical and surgical hospital wards, despite these services being routinely provided internationally. The aim of this study was to understand the impact of removing weekend allied health services from acute medical and surgical wards using a disinvestment-specific non-inferiority research design. Methods and findings: We conducted 2 stepped-wedge cluster randomised controlled trials between 1 February 2014 and 30 April 2015 among patients on 12 acute medical or surgical hospital wards spread across 2 hospitals. The hospitals involved were 2 metropolitan teaching hospitals in Melbourne, Australia. Data from n = 14,834 patients were collected for inclusion in Trial 1, and n = 12,674 in Trial 2. Trial 1 was a disinvestment-specific non-inferiority stepped-wedge trial where the ‘current’ weekend allied health service was incrementally removed from participating wards each calendar month, in a random order, while Trial 2 used a conventional non-inferiority stepped-wedge design, where a ‘newly developed’ service was incrementally reinstated on the same wards as in Trial 1. Primary outcome measures were patient length of stay (proportion staying longer than expected and mean length of stay), the proportion of patients experiencing any adverse event, and the proportion with an unplanned readmission within 28 days of discharge. The ‘no weekend allied health service’ condition was considered to be not inferior if the 95% CIs of the differences between this condition and the condition with weekend allied health service delivery were below a 2% increase in the proportion of patients who stayed in hospital longer than expected, a 2% increase in the proportion who had an unplanned readmission within 28 days, a 2% increase in the proportion who had any adverse event, and a 1-day increase in the mean length of stay. The current weekend allied health service included physical therapy, occupational therapy, speech therapy, dietetics, social work, and allied health assistant services in line with usual care at the participating sites. The newly developed weekend allied health service allowed managers at each site to reprioritise tasks being performed and the balance of hours provided by each professional group and on which days they were provided. Analyses conducted on an intention-to-treat basis demonstrated that there was no estimated effect size difference between groups in the proportion of patients staying longer than expected (weekend versus no weekend; estimated effect size difference [95% CI], p-value) in Trial 1 (0.40 versus 0.38; estimated effect size difference 0.01 [−0.01 to 0.04], p = 0.31, CI was both above and below non-inferiority margin), but the proportion staying longer than expected was greater with the newly developed service compared to its no weekend service control condition (0.39 versus 0.40; estimated effect size difference 0.02 [0.01 to 0.04], p = 0.04, CI was completely below non-inferiority margin) in Trial 2. Trial 1 and 2 findings were discordant for the mean length of stay outcome (Trial 1: 5.5 versus 6.3 days; estimated effect size difference 1.3 days [0.9 to 1.8], p < 0.001, CI was both above and below non-inferiority margin; Trial 2: 5.9 versus 5.0 days; estimated effect size difference −1.6 days [−2.0 to −1.1], p < 0.001, CI was completely below non-inferiority margin). There was no difference between conditions for the proportion who had an unplanned readmission within 28 days in either trial (Trial 1: 0.01 [−0.01 to 0.03], p = 0.18, CI was both above and below non-inferiority margin; Trial 2: −0.01 [−0.02 to 0.01], p = 0.62, CI completely below non-inferiority margin). There was no difference between conditions in the proportion of patients who experienced any adverse event in Trial 1 (0.01 [−0.01 to 0.03], p = 0.33, CI was both above and below non-inferiority margin), but a lower proportion of patients had an adverse event in Trial 2 when exposed to the no weekend allied health condition (−0.03 [−0.05 to −0.004], p = 0.02, CI completely below non-inferiority margin). Limitations of this research were that 1 of the trial wards was closed by the healthcare provider after Trial 1 and could not be included in Trial 2, and that both withdrawing the current weekend allied health service model and installing a new one may have led to an accommodation period for staff to adapt to the new service settings. Stepped-wedge trials are potentially susceptible to bias from naturally occurring change over time at the service level; however, this was adjusted for in our analyses. Conclusions: In Trial 1, criteria to say that the no weekend allied health condition was non-inferior to current weekend allied health condition were not met, while neither the no weekend nor current weekend allied health condition demonstrated superiority. In Trial 2, the no weekend allied health condition was non-inferior to the newly developed weekend allied health condition across all primary outcomes, and superior for the outcomes proportion of patients staying longer than expected, proportion experiencing any adverse event, and mean length of stay.