An impact evaluation of an education bundle for patients at risk of developing venous thromboembolism

The model follows a single cohort over a 5-year time horizon. The cohort represents the hospitalized US population at risk for developing VTE. An estimated 12 million patients a year experience a procedure or medical event during hospitalization that puts them at risk of VTE [15]; these 12 million patients cycle through the model and accumulate costs and outcomes. These 12 million patients were split into two risk groups of 4.3 million surgical patients and 7.7 million medical patients [15]. Patients in these risk groups were subsequently assigned differing VTE risk estimates, following the literature. Because the study was modeling patients at increased risk of VTE, more potential VTEs could be prevented by the intervention than would be present in the general hospitalized population. Patients entered at age 55 years, which was the average age of patients participating in the VTE education bundle study [10].

For patients who received the intervention, the average reduction in pharmacologic missed doses was calculated as 0.632 doses per patient using estimates from Haut and colleagues [10], and the cost of the additional dose per participant receiving the intervention was modeled. The decrease in VTE events as measured by administrative codes (PSI-12, Perioperative Pulmonary Embolism or Deep Vein Thrombosis Rate) observed by Haut and colleagues [10] was not found to be statistically significant.

Although Haut and colleagues [10] were significantly powered to observe a statistically significant change in the rate of prophylaxis administration, the study was underpowered to observe a significant decline in VTEs within and across VTE risk groups, and few VTEs were observed in the trial. Louis and colleagues found a statistically significant association between missed doses of prophylaxis and VTEs among trauma and general surgery patients, in which 23.5% of trauma and general surgery patients who missed at least one dose of prophylaxis experienced a VTE compared with 4.8% of patients who missed zero doses [6]. Estimates from Louis and colleagues [6] were applied to the 4.3 million general surgery patients in the at-risk for VTE population, and estimates from Haut and colleagues [10] were applied to the rest of the at risk for VTE population to estimate the rates of patients experiencing VTEs across risk groups in the intervention and control. Total hospital days, VTE events and VTE follow-up visits made up the remaining outcome parameters. A shorter length of stay (LOS) was modeled for patients not experiencing VTE during initial hospitalization. Patients experience an estimated 2.89 additional follow-up visits per VTE event [16]. Supplementary Table 1 provides the full list of model parameters.

Patients can die at any point as they transition through the model. Background mortality was applied to all health states, and event-specific mortality was applied to the relevant health state. Event-specific mortality rates included a general hospital-related rate (for initial hospitalization that did not result in VTE), VTE-specific rate (for all initial hospitalization VTEs and future VTEs) and increased risk of mortality following VTE applied to the post-VTE health state.

Direct medical costs include cost of hospitalization, post-VTE medical costs, and cost of the additional pharmacologic prophylaxis dose in the intervention arm. Out-of-pocket costs for inpatient stays were estimated by creating an out-of-pocket cost to total cost ratio, constructed from the ratio of total costs to out-of-pocket costs for all medical events in the Medical Expenditure Panel Survey (MEPS). This ratio was multiplied by the total cost of inpatient stays found by Cohoon and colleagues [17].

The original Haut and colleagues study [10] had a dedicated health educator who delivered the intervention. In a later dissemination study [18], all floor nurses received training and delivered the intervention, because most hospitals would not hire a new health educator just to deliver this intervention. It was assumed that removing the dedicated health educator did not affect the observed decrease in VTE events used in the model. The responsibility of delivering the intervention was assumed to be split by all nurses providing care for the at-risk population (see all other assumptions in Supplementary Table 2). Nurses would be required to undergo a one-time, 30 min online training to deliver this intervention. Material costs modeled included the cost of printing the paper handout given as an option for the education bundle. The other options for providing the intervention were assumed not to incur any costs because face-to-face discussion was assumed to be a standard nurse responsibility and the educational video was made freely available on the internet. Intervention costs include nurses’ wages (for the time spent in training), the cost of the online training, and the paper and printing costs for the education bundle. The cost of the training module for providers was estimated to be approximately US$11 per nurse. Other intervention costs included the cost of implementing the real-time alert system in hospitals. These costs were calculated based on author’s assumptions of time to implement the real-time alert system at the wage of a software developer [19]. Maintenance costs were also included for each subsequent year. These costs were assumed across all US hospitals [20].

Patient time, travel and employer productivity cost parameters were estimated between the control and intervention arms. Patient time included time spent receiving VTE-related medical care, including time in the hospital and follow-up visits. Patient total time and absenteeism time were estimated from average LOS for VTE and non-VTE hospitalization. Absenteeism time was estimated as a 5-day workweek of 40 h. Productivity loss to employers was estimated by multiplying the labor force participation rate among Americans aged 55 years and older [10] by the estimated absenteeism hours resulting from hospitalization and follow-up visits and an average patient wage of US$35.96 [21]. A similar estimate was created for productivity losses to patients’ families and friends serving as caregivers. The caregiver wage rate used was an estimate of the median wage for nursing and home health aides [21]. Travel costs were estimated using a nationwide median distance between patients’ homes and the closest hospital [22], a standard mileage rate of 59 cents per mile for 2020 from the Internal Revenue Service [23] and the number of follow-up visits for patients who experience VTE (estimated from the work of Spyropoulos and Lin [16]). Please see Supplementary Table 1 for a table of all parameter values and their sources.

To explore uncertainty in the model estimates, one-way analysis and probabilistic sensitivity analysis (PSA) were performed. One-way sensitivity analysis involved varying each model input, one at a time, to determine how each input affected total costs. The results are presented graphically as a tornado diagram identifying primary cost drivers in the model. PSA was conducted to allow for uncertainty in the estimation of all parameters within the model. Each input parameter was assigned a probability distribution and a range for which the default input would vary in PSA. The range was informed by 95% CIs found in the literature, a credible range, or a 20% increase and decrease in the default input. PSA jointly varied model inputs over 1000 simulations from these distributions to estimate the average VTE events, hospital days, mortality, medical costs and nonmedical costs for each group. Then, 95% credible range estimates were calculated for each model outcome based on these results. Additional analyses examined time horizons from 1 to 10 years.