Cost-Effectiveness of Universal Serological Screening to Prevent Non-Traumatic Hip and Vertebral Fractures in Patients with Celiac Disease

Decision-analytic Model

We constructed a decision-analytic Markov model of 12 year old cohorts with population-based prevalence of CD in North America. A natural history and progression toward hip bone and vertebral fractures were used as clinical endpoints to assess the cost-effectiveness of providing universal CD serologic screening. A base case age of 12 years was determined clinically relevant for serologic CD testing since dietary habits are more likely shaped by peers, and primary physicians are screening for baseline anemia and dyslipidemia as per standard of care during preadolescence.^22,23 Because the natural age progression to osteoporosis and bone loss is different for male and female, our model is categorized in two groups based on male or female gender. We considered 2 strategies for comparison, as shown in Figure 1: universal serologic screening versus standard of care screening.

Our model followed the recommendations of the US Panel on Cost-Effectiveness in Health and Medicine in the development and the analysis of results, using a societal perspective, considering costs and benefits over a lifetime horizon, and discounting at 3% annually.²⁴ Base case parameter estimates and ranges and distributions used in the sensitivity analysis are presented in Table 1. We constructed and implemented the model in TreeAge Pro 2012 Suite (TreeAge Sofware Inc, Williamstown, MA) and Microsoft Excel 2007 (Microsoft Corportation, Redmond, WA). The two cohorts progressed through the model in 1-year time steps until death or 100 years of age. Within each health state, patients could die at a rate based on the average age-specific mortality tables, as estimated by the Centers for Disease Control and Prevention.²⁵ The model analyzed the differences in lifetime discounted costs and benefits (measured in life-years and quality-adjusted life years) between the 2 competing strategies. The primary outcome measure was the incremental cost-effectiveness ratio (ICER), which is defined as the difference in costs in dollars divided by the difference in effectiveness in quality-adjusted life-years (QALYs) between the 2 competing strategies.

Symptomatic and At-Risk Screening (SAS)

The current standard of care guidelines for serologic screening captures at-risk populations and patients with symptomatic gastrointestinal complaints for CD. Populations at-risk for CD, as defined by literature, include Type 1 diabetes mellitus, Down, Turner, and Williams syndromes, immunoglobulin A deficiency, systemic lupus, autoimmune thyroiditis, and having a first and/or second degree relative with CD (Table 1).^{422,43,44,45,46,47,48,49,50,51,52,53} Under this strategy, patients without a genetic or hereditary risk factor for CD were normally screened due to associated gastrointestinal complaints, such as diarrhea, abdominal pain, bloating, and other irritable-bowel-like symptoms.^34,35,36,37 In children, poor growth, wasting, failure-to-thrive, and anemia also warranted CD screening (Table 1).^4,17,26 Any positive serologic screens required diagnostic confirmation via endoscopic duodenal mucosal biopsies. Once a CD patient started life-long therapy by maintaining a strict gluten-free diet (GFD), patients were subject to natural adherence and non-adherence rates reported in literature.^54,55,56,57

Universal Serologic Screening (USS)

Under this strategy, all patients received a one-time serologic test for immunoglobulin A antibodies to tissue transglutaminase (tTG IgA) and total immunoglobulin A (IgA). The combination of high levels of tTG IgA with normal IgA is validated as a screening modality of choice, conferring high sensitivity and specificity in subjects >2 years of age (Table 1).¹⁷ As in the SAS strategy, all positive screening tests were then confirmed via duodenal mucosal biopsies. The natural probabilities of adherence and non-adherence to strict GFD were also applied to this cohort after diagnosis of CD (Table 1).

Hip & Vertebral Fractures

Current literature supports the development of deteriorating bone disease among undiagnosed CD patients and CD patients who were poorly managed or non-adherent to strict GFD (Table 1). We assumed that CD patients with progressive bone disease developed non-traumatic fractures at the same rate as the non-CD population who had comparable bone demineralization. The model aims specifically to evaluate the health and economic consequence of long bone (i.e., hip) and vertebral fractures among CD patients who are undiagnosed or untreated with GFD. We chose to focus on the development of these two types of fractures in particular in order to ensure that – whenever possible – the model captures the specific costs and quality-of-life trade-offs between the competing screening strategies based on clearly defined clinical end-points. Hip and vertebral fractures also carry the highest morbidity rate in terms of progressing to long-term disability.^32,33

Disability

Patients in the model who developed hip or vertebral fractures were at risk for developing long-term disability. The surgical success rates of total hip arthroplasty and kyphoplasty were included in the model.^58,59 Among patients who require surgery, if they did not achieve full functional capacity after a 1 year cycle, patients incurred decrements in quality of life and medical disability costs as reported by literature in a disabled state, where they remained until death.

Costs

Cost estimates for total hip arthroplasty, kyphoplasty, and annual disability after fractures were derived from the Centers for Medicare and Medicaid Services (CMS) for 2012.⁶³ Additional costs required in CD for a strict GFD were approximated by considering the report by Lee et al⁶¹ and the baseline average monthly costs for food among males and females according to the USDA.⁶² We assumed bi-annual physician follow up clinic visits and one initial educational visit with a dietician for CD. These added healthcare provider costs were estimated by querying de-identified patient billing records in 2011 from Lucile Packard Children’s Hospital (LPCH) and Stanford University Medical Center (SUMC). The cost of serologic screening using tTG IgA and total IgA was reported by Shamir et al⁶⁰ and validated by the clinical laboratory at LPCH and SUMC.

Utilities

Utilities estimate patient quality of life (QOL), ranging from 0 (death) to 1 (perfect health). Previous literature suggested a difference in QOL between symptomatic and asymptomatic CD. The utility of 0.71 for symptomatic CD (i.e., abdominal pain, bloating, diarrhea, etc) in Table 1 was a conservative estimate based on two utility estimates. We averaged 0.66 based on a report from Norström et al⁶⁴ surveying CD patients 1 year prior to CD diagnosis and 0.76 based on QOL living with dyspepsia and diarrhea in Tengs and Wallace.⁶⁵ Our estimation of 0.71 was validated in a cost-effectiveness analysis performed by Hershcovici et al.⁶⁶ Patients who recovered completely after a fracture did not progress to the disabled state and were subject to same pre-fracture transition probabilities. For disabled patients after a fracture, we incorporated the reports from Jönsson et al⁶⁷ and Oleksik et al⁶⁸ that suggest patients who had either a hip or vertebral fracture exhibit a QOL that is approximately 90% of the previous non-fracture or healthy state. The QOL in disability varies considerably, and our approximation may not adequately represent poor QOL in severely disabled states. However, we critically evaluated this variable under a wide sensitivity range to ensure that a substantially lower QOL in disability does not impact the results.

Sensitivity Analyses

We performed a deterministic and a probabilistic sensitivity analyses on all health state probabilities, costs, and utilities in the model (Table 1). One-way deterministic sensitivity analyses were performed to identify variables that had important effects on the decision for either USS or SAS over the specified range of values reported in literature or expert opinion. Each variable was tested separately to determine whether varying the particular variable over a broad range alters the ICER. For probabilistic sensitivity analysis and for calculating the 95% confidence intervals around the base case results, we performed 1,000 independent Monte Carlo simulations. Beta distributions were used for probabilities and utilities. Lognormal distributions were used for costs. We assumed a willingness-to-pay (WTP) threshold of <$50,000 per QALY as being cost-effective, but generated acceptability curves for each strategy over a $0 to $100,000 WTP thresholds to estimate the net benefit.

Symptomatic and At-Risk Screening (SAS) vs. Universal Serologic Screening (USS)

Our model shows complete dominance of the USS strategy by the SAS strategy, which is the current standard of care in most clinical practices. The USS strategy for either gender is not cost-effective. Table 2 shows a summary of the cost-effectiveness of the 2 strategies. For males, USS accrues a life-time average cost of $8,532 with an associated QALY-gained of 25.511, but SAS had lower costs of $8,472 and minimally higher QALY-gained of 25.515. Similarly, for females, USS accrues a life-time average cost of $11,383 with an associated QALY-gained of 25.74, while SAS had lower costs of $11,328 and minimally higher QALY-gained of 25.75. Figure 2 shows the cost-effectiveness planes of the two strategies stratified by gender.

Deterministic Sensitivity Analyses

We tested the robustness of our model by varying each of the variables over a wide range of values as listed in Table 1. This method identifies the key parameters that impact the cost-effectiveness of the competing strategies. One-way sensitivity analysis identifies potential variables that may change the dominance of one strategy to a competing strategy, if base case values are different than what is assumed in the model. No variable was found to be significant, and potential changes in the individual variables did not change the ICER for either males or females. In other words, we did not identify any specific threshold value for any of the variables whereby changing the base value within sensitivity range limits would alter the result of the model from a cost-ineffective to a cost-effective USS strategy. From an epidemiologic perspective, we hypothesized that a high prevalence of celiac disease in the general population could conceivably make the USS strategy more cost-effective. However, our model indicates that with any CD prevalence > 1.03%, the average QALY gained in the SAS strategy was greater than in the more expensive USS strategy. The implication is that false positives by serology screening would drive unnecessary costs and invasive endoscopy, which carries a separate and non-negligible risk.²⁷

Probabilistic Sensitivity Analysis

Probabilistic sensitivity analysis using multivariable Monte Carlo simulation of the model supports the dominance SAS as the more optimal strategy in detecting CD. We performed 1,000 independent simulations of the model (Figure 3). SAS strategy is more cost-effective in 67-80% of the 1,000 iterations of the model for men and women within all WTP thresholds < $100,000 per QALY-gained. Even at higher WTP thresholds between $50,000 to $100,000 per QALY-gained, USS strategy accrues more expense and yields less lifetime QALYs in approximately 70% of the time for either men or women. In every clinically realistic scenario when each variable is allowed to fluctuate between individual 95% confidence intervals, the current SAS strategy remains a more cost-effective approach to serologic screening for CD. For either gender, the likely clinical outcome under SAS strategy is increased QALYs and decreased costs (Figure 4), as indicated on an incremental cost versus incremental QALY plot.