The Specialist Treatment of Inpatients: Caring for Diabetes in Surgery (STOIC-D Surgery) Trial: A Randomized Controlled Trial of Early Intervention With an Electronic Specialist-Led Model of Diabetes Care

Hyperglycemia and hypoglycemia occurring in-hospital is known to be associated with increased mortality (1,2). Trials have focused on treating and preventing hyperglycemia in high-acuity inpatients, with variable evidence of benefit. One approach in the intensive care unit (ICU) setting has been to target a specific glucose range. In 2001, van den Berghe et al. (3) reported a mortality benefit in patients admitted to a surgical critical care unit managed with intensive rather than conventional blood glucose (BG) targets. In contrast, in a larger multicenter critical care study, the Normoglycemia in Intensive Care Evaluation–Survival Using Glucose Algorithm Regulation (NICE-SUGAR), investigators found similar intensive BG targets conferred increased 90-day mortality, while the follow-up Tight Blood-Glucose Control (TGC)-Fast trial found that tight glucose control does not affect mortality or time to ICU discharge (4,5).

Another approach has been to prescribe a specific therapy. The Randomized Study of Basal-Bolus Insulin Therapy in Type 2 Diabetes in Surgery (RABBIT-2 Surgery) trial found basal-bolus insulin therapy resulted in reductions in both mean glucose and a composite of postoperative complications compared with sliding scale insulin, though at the cost of greater hypoglycemia (6). Evidence has been mixed for intensive (versus conventional) insulin therapy in the post–acute myocardial infarction setting, where the Diabetes Mellitus Insulin Glucose Infusion in Acute Myocardial Infarction 1 (DIGAMI-1) trial found that intensive insulin therapy resulted in reduced 1-year mortality, a result not seen in the DIGAMI-2 trial (7,8). In a before-after implementation trial in patients with diabetes undergoing cardiac surgery, perioperative continuous insulin infusion reduced the deep sternal wound infection (DSWI) rate compared with intermittent subcutaneous insulin injections (9).

A third approach has involved the implementation of specialist diabetes teams, which proactively identify patients with diabetes or adverse glycemia and provide advice or intervene without a referral from the treating team. The virtual glucose management service involved a clinician placing an electronic medical record (EMR) note with insulin change recommendations for inpatients identified to have ≥2 BG values of ≥12.5 mmol/L (10). It was associated with a reduction in adverse glycemia. Similarly, in the cluster-randomized Randomized Study of a Proactive Inpatient Diabetes Service (RAPIDS) trial of early specialist bedside consultation and individualized management of non-ICU patients, early specialist intervention reduced glucose and health care–associated infections (HAIs) (11).

Limitations of previous trials include a focus on high-acuity patients in critical and cardiac care settings to the exclusion of general ward patients, use of a one-size-fits-all medication treatment protocol, requirements for substantial trial teams, and evidence generation in pre-EMR contexts. To address these limitations, the Specialist Treatment of Inpatients: Caring for Diabetes (STOIC-D) trial program was conceived. In this first STOIC-D randomized clinical trial in a surgical setting, we investigate the effect of early intervention with an electronic-based specialist-led “proactive” model of care on glycemic and clinical outcomes.

The STOIC-D Surgery trial was a large, single-center, parallel group, randomized, controlled trial performed at the Royal Melbourne Hospital, a quaternary referral hospital affiliated with the University of Melbourne in Australia. The trial protocol and all subsequent amendments (Supplementary Appendix 1, pp. 1–34) were approved by the ethics committee of The Royal Melbourne Hospital (Melbourne, Australia; HREC/61095/mH-2020). A waiver of individual consent was granted on the basis of the study design and procedures conforming to section 2.3.10 of the National Health and Medical Research Council Statement on Ethical Conduct in Human Research – 2007 (Updated 2018) (12). The trial was registered in the Australian New Zealand Clinical Trials Registry (ACTRN12620001303932).

All nonpregnant adult patients admitted to a long-stay ward of the study center’s acute campus under a surgical inpatient unit and identified as having either a diagnosis of diabetes (excluding type 1 diabetes; these patients were assessed in a dedicated trial) made prior to the admission or newly detected hyperglycemia, defined as at least one random BG result ≥11.1 mmol/L recorded during the admission, were eligible for the study. All eligible participants were recruited and randomized at the time of identification (trial entry). BG result identification was assisted by the use of networked BG meters (StatStrip; Australasian Medical & Scientific, Ltd., Nova Biomedical), which transmit point-of-care capillary glucose and ketone results to the EMR in real time.

Participants contributed data to the study until trial exit. Trial exit was defined as the occurrence of one of the following events: discharge from the acute campus of the study center, transfer to a nontrial ward (e.g., palliative care, psychiatry), transfer to any unit other than a surgical unit, or death. Participants contributed data while admitted to either general wards or the ICU.

If a participant experienced multiple unique admissions to the study center during the recruitment period and more than one admission met the inclusion criteria, all such admissions were included. Any mode of entry to the surgical inpatient unit was acceptable, for example, via the emergency department or interhospital transfer for urgent surgical treatment, or via admission on the day of surgery for a planned (elective) surgical procedure.

Participants were randomly assigned (1:1) by the investigators to receive early electronic proactive specialist diabetes care (intervention) or standard diabetes care (control) during their inpatient stay. Randomization was performed using a computer-generated random sequence (dynamic block randomization with blocks of size 4, 6, or 8) with an internet-based program. The specialist inpatient diabetes service was informed on the morning of each business day of new participants randomized to receive early electronic proactive diabetes care but was not informed of participants allocated to standard diabetes care. Participants were not informed of their treatment allocation. Adjudication of outcomes was performed by an investigator (L.J.W.) who was blinded to treatment allocation and had no involvement in trial procedures.

For participants allocated to standard diabetes care, diabetes management while an inpatient of the study institution was performed primarily by the hospital medical officers of the admitting unit who could access a specialist inpatient diabetes team (IDT) as required. An IDT consisting of a diabetes specialist nurse and endocrinology trainee (registrar) doctor supervised by a diabetologist provided a consultation service. This service involved the IDT performing a bedside diabetes consult (BDC) in response to referrals from the admitting unit. During a BDC, the IDT may directly make modifications to a participant’s inpatient diabetes management, typically in consultation with the admitting unit. A combination of insulin and noninsulin agents was used to achieve the participant’s individualized glycemic target, with the default target typically 5.0–10.0 mmol/L, and avoid adverse glycemia. Choice of agents and dosage depended upon glycemic targets, comorbidities, contraindications, and allergies, per accepted specialist practice. Noninsulin agents used included predominantly metformin and dipeptidyl peptidase 4 inhibitors. Sodium–glucose cotransporter 2 inhibitors and glucagon-like peptide 1 receptor agonists were used infrequently for inpatient glycemic management, and, when prescribed, this was usually prior to discharge for the purpose of modifying a comorbid patient’s ambulatory diabetes management. The IDT used consistent inpatient diabetes management principles including use of basal-bolus insulin regimens based on empiric calculations (initial total daily insulin dose 0.4 units/kg/day) for patients experiencing hyperglycemia. The IDT would involve the consultant diabetologist on an as required basis and discuss cases in weekly multidisciplinary team meetings. The IDT would remain involved in the participant’s diabetes care until their input was no longer clinically indicated and would then cease to review the participant. The admitting unit could subsequently rerefer to the IDT to reinvolve them in the participant’s care. Our institution had guidelines and protocols on inpatient diabetes management, including a perioperative diabetes management protocol (13), and an insulin order set available in the Epic (Epic Systems Corporation, Verona, WI) EMR. There was no general diabetes or glucose management education undertaken for the purposes of this study. Patients undergoing elective surgical procedures who were identified by the surgical team as being at risk for adversity were reviewed in the preanesthetic clinic in the weeks prior to their procedure. It was at this clinic that the perioperative diabetes management protocol, if relevant, was first enacted. Patients undergoing elective procedures are almost universally admitted on the day of the procedure. All patients are recommended to have at least one glucose measure on hospital admission. Patients with diabetes and newly detected hyperglycemia are recommended to have daily four-point capillary glucose monitoring (premeal three times per day and prebed).

For participants allocated to early intervention with electronic specialist-led proactive diabetes care, after randomization, the following was delivered by an endocrinology fellow: 1) In an electronic diabetes consult (EDC), using a standardized consult template (Supplementary Appendix 2, pp. 2–3), diabetes management recommendations were recorded in the EMR for the admitting unit as a consultation note within one business day of randomization. The endocrinology fellow made recommendations only and did not directly change management during the EDC. Ongoing daily electronic review of participants occurred for the duration of the hospital stay. 2) In electronic message notification to the admitting surgical unit, the availability of initial EDC recommendations was additionally directly electronically communicated to the surgical unit hospital medical officer caring for the trial participant via the “Secure Chat” function in the Epic EMR. 3) For electronic escalation to the IDT, if certain criteria indicating a higher likelihood of adverse glycemia were present (Supplementary Appendix 2, p. 4), escalation occurred to the IDT, who would perform a BDC on the same business day without a referral from the admitting unit, that is, a proactive consult. BDCs performed on participants allocated to the intervention group were identical to those performed on participants allocated to standard care (see description of BDC procedures above). The IDT would remain involved in the participant’s diabetes care until their input was no longer clinically indicated and would then cease to review the participant. Ongoing daily electronic review of participants not currently escalated to BDC continued, and, if criteria for escalation were subsequently met, the IDT would be informed at the time and perform a BDC.

A graphical representation of trial procedures is shown in Supplementary Appendix 1 (p. 17).

The primary outcome was patient-day mean glucose (PDMG), calculated by grouping all glucose results by patient-day and returning a mean value for each patient-day (14). Key secondary outcomes included HAI (defined as a sterile site positive culture or clinician action [antibiotic prescription] on a suspected infection where both the positive culture was taken and the antibiotic prescription commenced at least 48 h following admission), and proportion of participants with patient-day–weighted mean glucose <12.0 mmol/L (defined as the mean of all PDMG values for that participant over their inpatient stay).

Exploratory secondary outcomes included process, glucometric, and clinical outcomes, and are listed in Supplementary Appendix 1 (pp. 11–12). Safety outcomes were mortality and hypoglycemia.

Patient-day glucose mean and SD estimates were obtained from Kyi et al. (11) To detect a difference in PDMG, group sample sizes of 1,102 (551 participants per treatment group) would achieve 80% power to reject the null hypothesis of equal means when the population mean difference is −0.5 with SDs of 2.7 for group 1 and 3.2 for group 2, and with a significance level (α) of 0.05 using a two-sided two-sample unequal variance t test. Assuming a 10% drop-out rate, 1,226 participants (613 per treatment group) needed to be recruited.

Baseline and outcome data were summarized separately. Continuous variables were summarized as mean (SD) or median (interquartile range [IQR]) depending on the variable’s distribution. Dichotomous variables were reported as percentages. Continuous outcomes were assessed using the t test if normally distributed, and Wilcoxon rank sum test if not. Dichotomous outcomes were analyzed using the χ² test. Differences and CIs in variables with nonnormal distributions were obtained using the Hodges-Lehmann estimator. BG measurements were excluded after day 14 of admission to avoid undue influence by prolonged hospital stays upon summary statistics.

Efficacy and safety analyses were performed according to the intention-to-treat principle and included all randomly assigned participants. We compared the rates of the primary outcome (i.e., PDMG) between the intervention and control group. Differences in PDMG were expressed as the absolute difference in millimoles per liter. For secondary outcomes, risk ratios were calculated.

Statistical analysis was performed using R version 4.0.3 (R Foundation for Statistical Computing, Vienna, Austria). There was no data monitoring committee. The trial was registered in the Australian New Zealand Clinical Trials Registry (ACTRN12620001303932).

Data pertaining to this study, including individual participant data, are held in a deidentified form by The Royal Melbourne Hospital. These data will be available for a minimum of 7 years following publication of this report with investigator support to individuals or groups submitting a proposal to the Office for Research that is approved, after receipt of a signed data-sharing agreement. Data and documents, including the study protocol, will be provided by secure electronic transfer.

Between 12 February 2021, and 17 December 2021, of the 1,417 admissions assessed for study eligibility, 1,371 met inclusion criteria and were randomly assigned to early intervention with electronic specialist-led diabetes care (680) or standard diabetes care (691) (Fig. 1). Baseline characteristics were well balanced between the intervention and control arms (Table 1). In the intervention and control groups, mean age was 64.6 ± 15.8 years and 65.4 ± 16.0 years, respectively, 426 (63%) and 433 (63%), respectively, were male, and 476 (70%) and 480 (69%), respectively, had diabetes. The median modified Charlson comorbidity index was 2 in both groups. There were 96 patients who contributed more than a single admission to the trial.

The early intervention group achieved a lower median PDMG of 8.2 mmol/L (IQR 6.9–10.0 mmol/L) compared with 8.6 mmol/L (IQR 7.2–10.3 mmol/L) in the control group (Fig. 2). The estimated difference in PDMG between groups is −0.3 mmol/L (SD 0.05 mmol/L, 95% CI −0.4 to −0.2 mmol/L, P < 0.0001).

The incidence of HAIs was lower in the intervention compared with the control group (77 [11%] vs. 110 [16%], respectively), which corresponds to an absolute risk difference of −4.6% (95% CI −8.2 to −1.0, P = 0.016) (Fig. 3). The number needed to treat for HAI prevention was 22 inpatients. The most common site of infection was the respiratory tract, accounting for 104 infections (56% of all HAIs). Other key secondary outcomes were no different between the intervention and control groups (Supplementary Appendix 2, p. 5).

Adverse events, including hypoglycemia, defined as a BG of <4.0, <3.0, or <2.2 mmol/L, and mortality were no different between groups (Supplementary Appendix 2, p. 6). Fatal traumatic injuries were the most common cause of mortality, accounting for 22 (49%) of the 45 deaths. Causes of mortality are described in Supplementary Appendix 2 (p. 6).

All 680 patients randomized to the intervention received an EDC on the day of randomization. The median difference between the date of admission and of the first EDC in the intervention group was 1 day. A bedside consultation was performed at least once in 333 patients (49%) of the intervention and 93 patients (14%) of the control group. Among those patients seen at the bedside, there was a median difference of 0 days both between the first EDC and first BDC in the intervention group and between referral to the IDT and the first BDC in the control group. That is, the first BDC most often occurred on the same day as the initial escalation to the IDT, either proactively through the intervention or in response to a referral from the admitting unit. For those patients seen at the bedside, the median difference between the date of admission and of the first BDC was 2 days for the intervention group and 4 days for the control group. Of the 348 patients in the intervention arm seen exclusively electronically, 168 (48%) received more than one EDC.

Regarding surgical management choices that could affect glucose postoperatively, there was no difference in the proportion of patients administered perioperative dexamethasone in the intervention group (170 [25%]) compared with the control group (176 [25%]). Given the typical practice at our institution of avoiding the use of preoperative glucose drinks in patients with diabetes or hyperglycemia, we did not identify any trial patients having received such a drink. There was, similarly, no difference in the proportion of patients receiving enteral or parenteral nutrition between the intervention (103 [15%]) and control (111 [16%]) groups. Of the patients in the control group receiving supplemental nutrition, 15 (14%) were seen at the bedside by the IDT, while all of the intervention group patients were seen. There were no differences between the groups in ICU admission or intravenous insulin use.

STOIC-D Surgery, to our knowledge, is the largest whole-of-hospital randomized controlled trial of a diabetes model-of-care intervention to date, including glucose and outcome data from both critical and noncritical care locations. The trial achieved its primary outcome of a reduction in PDMG and the key secondary outcome of reduced HAIs. This establishes the effectiveness of dedicated early intervention with an electronic specialist-led diabetes model of care to reduce morbidity in hospital.

Delivery of proactive diabetes care, a major component of the STOIC-D trial intervention, has previously been associated with reductions in adverse glycemia in before-after (10,15–19) and randomized controlled trial (11,20,21) settings. While the form of proactivity implemented differs between studies, conserved features include identification of patients with or at risk for adverse glycemia by a clinician or team with expertise in inpatient diabetes management, followed by involvement in the patient’s management at an earlier time point than would have otherwise occurred in the context of that institution’s contemporary practices. STOIC-D Surgery also represents the first clinical trial of a diabetes model of care randomized at the level of the individual, rather than admitting unit or ward.

That a reduction in HAIs was observed following a modest 0.3 mmol/L reduction in PDMG is consistent with previous evidence. The Portland studies showed perioperative insulin infusion therapy to reduce glucose post coronary artery graft surgery in patients with diabetes reduces the incidence of DSWI from 2.0% to 0.8% compared with subcutaneous insulin injections (9,22). Those with DSWI had a reduction of 0.9 mmol/L (mean glucose of 11.5 vs. 10.6 mmol/L in those without). The RAPIDS cluster-randomized trial of early specialist intervention for all patients with diabetes achieved a 4.0% absolute and 60% relative risk reduction in HAIs (11) with a 0.4 mmol/L reduction in PDMG. Following withdrawal of the intervention, both hyperglycemia and HAIs returned to their preintervention rates, strongly suggesting an etiologic link between the two (23). A meta-analysis in noncritical care settings similarly found intensive glycemic control to be associated with reduced HAI rates (24).

Considering standard BG monitoring practices, the PDMG on a typical patient-day will represent the average of the fasting (prebreakfast), prelunch, predinner, and prebed BGs. Assuming a patient is eating regular meals at standard times, all but the prebed level are local minimum values when considering the 24-h glucose profile and, in the case of the fasting BG, the absolute minimum value. Therefore, the PDMG, as a consequence of not reflecting prandial glucose excursions, typically represents an underestimate of the 24-h glycemic exposure. Thus, relatively small differences in PDMG likely represent comparatively larger differences in total glycemic exposure.

There is strong mechanistic evidence for a deleterious effect of hyperglycemia on immune function, in particular on innate immunity, which predisposes to HAI. Hyperglycemia has been associated with impaired polymorphonuclear leukocyte chemotaxis (25), reduced superoxide radical production (26), and inhibited neutrophil degranulation (27), which all result in impaired microbial neutralization and elimination. Once infection has occurred, there is evidence that hyperglycemia increases in-hospital mortality, particularly in pneumonia (28). There are significant health care costs associated with HAIs. For example, in Australia, deep surgical site infections cost, on average, $13,187 Australian dollars (∼$8,500 U.S. dollars [USD]) per case in 2013 (29), while, in the U.K., vascular surgical site infections of all types cost £3,776 ($4,600 USD) in 2020 (30). If infection involves a prosthetic joint, costs can be as high as $70,000 Australian dollars (∼$44,600 USD) per case. In the STOIC-D Surgery intervention group, the number of inpatients needed to treat to prevent one in-hospital HAI was 22. The study ran for 309 days, with an absolute reduction of 33 HAIs (or one fewer HAI every 9 days) in the intervention groups. Substantial cost savings may thus be possible through the intervention reducing HAI incidence.

There is increasing recognition that noninsulin medications can be safe and effective in an inpatient setting (31). However, the considerations around these medications, including side effect profiles and contraindications, are often more involved than for insulin and require information from the patient not present in the clinical record to establish. In this way, a BDC, through acquiring this information, allows therapeutic individualization with appropriate noninsulin medication prescription, though the proportion of bedside consults during which proactive orders were placed was not assessed. This direct specialist-patient interaction allows consideration of the postdischarge diabetes management plan, enabling the admission to be an opportunity to optimize ambulatory care (32). While diabetes models of care involving predominantly electronic care (10) and bedside care (11) have been shown to be effective, most centers typically use only one. This trial hybridizes these models of care, leveraging the advantages of each to improve glucose and adverse inpatient outcomes.

Strengths of our trial include its large size and individual-level randomization, compared with previous largely before-after studies of proactive models of diabetes care. Glycemic management in our trial was based on a consistent team approach to use inpatient diabetes management principles with the capacity to individualize management where required rather than always using a single glucose target range (3,4) or a single glycemic medication approach (6–8). Importantly, this therapeutic individualization allowed hyperglycemia to be reduced without a concomitant increase in hypoglycemia or other adverse outcomes. The trial leveraged an EMR and networked BG meter technology, which enabled trial completion with a lean core clinical team consisting of a consultant endocrinologist, fellow, registrar, and nurse practitioner, all of whom concurrently performed their usual clinical roles. The early electronic proactive model of care has thus been shown to require minimal additional personnel resourcing for implementation, following implementation of prerequisite technologies. The STOIC-D Surgery intervention was enabled by the technology combination of an EMR and a fully automated EMR-integrated networked BG meter system, which was not present at any other Australian institution at the time of trial commencement.

Limitations of the study include generalizability. As the only Australian institution with both an EMR and networked BG meters at the time of trial commencement, this model of care could be applicable to institutions with these technologies. However, many institutions are planning or in the process of implementing these technologies such that this trial may represent an effective model of care to be adopted postimplementation. There was also prevalent coronavirus disease 2019 in the institution’s catchment area during the trial period (33), which resulted in some limitations to elective surgery. This would likely have modified the population enrolled, resulting in a greater proportion of acute surgical presentations. While the Hawthorne effect may have improved glucose management by admitting units and the IDT over the course of the trial, this would have reduced the glucose differences between the groups, with participants in the control group experiencing better glucose management over time.

Future studies could assess early electronic proactive models of diabetes care in different populations, institutional settings, and technological contexts. The criteria for escalation to BDC in the intervention group were relatively broad, resulting in 49% of these patients being seen at the bedside. There is thus opportunity to further refine these criteria to better target limited personnel resources. The clinical and cost effectiveness of this model of care needs to be assessed in multiple contexts including medical inpatients, and in diabetes types other than type 2.

In conclusion, STOIC-D Surgery is the largest whole-of-hospital diabetes randomized clinical trial to date and the first to individually randomize assessment of an early electronic specialist-led model of diabetes care. We have shown that early intervention with a technology-assisted specialist diabetes team can safely and effectively improve glucose across a large hospital inpatient population and, in so doing, reduce HAIs. Implementation of this proactive model of care in appropriate technology-enabled settings could significantly improve the care of and outcomes in people with diabetes and hyperglycemia admitted to hospital.

Funding. This investigator-initiated study was conducted with the support of the Rowe Family Foundation Perpetual Grant and The Royal Melbourne Hospital Home Lottery Victor Hurley grant. R.D.B. was supported by the Australian Government Research Training Program Scholarship.

The funders had no role in the study design, recruitment, data collection, analysis, interpretation, or writing of the report.

Duality of Interest. S.F. contributes to the advisory panel for Viatris Inc. and Pfizer Inc. S.F. contributed to the speaker’s bureau for Novo Nordisk, Astra Zeneca, and the Boehringer Ingelheim and Eli Lilly Alliance. No other potential conflicts of interest relevant to this article were reported.

Author Contributions. R.D.B., M.K., P.G.C., L.J.W., and S.F. were involved in the conception and design of the study. R.D.B., M.K., L.R., M.R., L.C., L.D., S.M., J.T., E.S., and M.L. were involved in the conduct of the study. R.D.B., M.K., and S.F. were involved in the analysis and interpretation of the results. R.D.B. wrote the first draft of the manuscript, and all authors edited, reviewed, and approved the final version of the manuscript. S.F. is the guarantor of this work and, as such, had full access to all the data in the study and takes responsibility for the integrity of the data and the accuracy of the data analysis.

Prior Presentation. This work was presented at the 83rd Scientific Sessions of the American Diabetes Association, San Diego, CA, 23–26 June 2023.

Handling Editors. The journal editors responsible for overseeing the review of the manuscript were Steven E. Kahn and Naveed Sattar.