The cost, survival, and quality‐of‐life implications of guideline‐discordant imaging for prostate cancer

Abstract Background National Comprehensive Cancer Network (NCCN) guidelines for incident prostate cancer staging imaging have been widely circulated and accepted as best practice since 1996. Despite these clear guidelines, wasteful and potentially harmful inappropriate imaging of men with prostate cancer remains prevalent. Aim To understand changing population‐level patterns of imaging among men with incident prostate cancer, we created a state‐transition microsimulation model based on existing literature and incident prostate cancer cases. Methods To create a cohort of patients, we identified incident prostate cancer cases from 2004 to 2009 that were diagnosed in men ages 65 and older from SEER. A microsimulation model allowed us to explore how this cohort's survival, quality of life, and Medicare costs would be impacted by making imaging consistent with guidelines. We conducted a probabilistic analysis as well as one‐way sensitivity analysis. Results When only imaging high‐risk men compared to the status quo, we found that the population rate of imaging dropped from 53 to 38% and average per‐person spending on imaging dropped from $236 to $157. The discounted and undiscounted incremental cost‐effectiveness ratios indicated that ideal upfront imaging reduced costs and slightly improved health outcomes compared with current practice patterns, that is, guideline‐concordant imaging was less costly and slightly more effective. Conclusion This study demonstrates the potential reduction in cost through the correction of inappropriate imaging practices. These findings highlight an opportunity within the healthcare system to reduce unnecessary costs and overtreatment through guideline adherence.


| INTRODUCTION
National Comprehensive Cancer Network (NCCN) guidelines for incident prostate cancer staging imaging have been widely circulated and accepted as best practice since 1996. 1 Despite these clear guidelines, inappropriate imaging of men with prostate cancer remains prevalent. 2 While imaging may yield useful additional information for some patients, for many imaging has a low probability of influencing treatment choices, and therefore, its harms outweigh its benefits. 3,4 Harms include emotional distress, wasteful spending, and increased cancer risk secondary to radiation exposure. On the other hand, underuse, or failure to provide appropriate imaging, can lead to delays in diagnosis, inadequate disease staging, and inappropriate treatment among patients who truly require care. The staging of incident prostate cancer is an ideal scenario to study appropriate imaging use, given the prevalence of guideline-discordant imaging and its cost to the healthcare system. To understand changing population-level patterns of imaging among men with incident prostate cancer, we created a simulation model based on existing literature and incident prostate cancer cases from the SEER-Medicare database. We believe a comprehensive understanding of changing imaging patterns, based on a nationally representative, adequately staged population, has the best chance to provide actionable suggestions for improvement of prostate cancer care.

| METHODS
Simulation modeling allows researchers to ask a series of what-if questions to understand how a population's life expectancy, quality of life, and health care spending could be impacted by changing how care is provided. When creating a simulation model, prior work has documented that transparency of the model aides is policy makers and clinicians. 5 We used data from SEER to analyze the survival for men using multivariable parametric proportional hazard survival models from the time   of diagnosis until death or censoring. Models were adjusted for age at   diagnosis as a categorical variable (65-70, 71-75, 76-80, 81-85, 85   +), stage at diagnosis (T1, T2, T2a, T2b, T2c, T3, and T4), PSA, and Gleason score. We determined the appropriate functional form (Weibull, Gompertz, or exponential) based on the models' Akaike Information Critera (AIC) and Bayesian information criterion (BIC), which indicated the Gompertz distribution was the best fitting model.

| Use of imaging
We used previously published observational research for individuals over age 65 in the United States, which documented use of imaging among incident prostate cancer cases. In incident high-risk prostate cancer cases in the SEER-Medicare data, only two-thirds of the population appropriately received imaging. 9 Concurrently, in incident low-risk prostate cancer cases approximately 45% of the population inappropriately received imaging. 9 In order to allow for correlation between patient factors and use of imaging, we generated the predicted probability of staging based on published odds ratios from a logistic regression. 11 We generated the constant of the logistic regression to ensure that the population staging was consistent with only two-thirds of high-risk men receiving appropriate imaging and 45% of low-risk men receiving inappropriate imaging.
Additionally, among low-risk men who received inappropriate imaging, prior research has shown that 38% received a false-positive test result.
Among these men, 43% received subsequent inappropriate imaging; we incorporated this into the model. 3

| Impact of imaging on treatment choice
A previous analysis of incident prostate cancer cases in SEER-Medicare data indicated that among men whose disease is metastatic, 94% will pursue systemic therapy, 5% radiation, and 1% surgery. 12 Among men who are high-risk but whose metastatic status is unknown or whose disease is known to be localized, prior research has found that 36% will pursue systemic therapy, 42% radiation, 6% surgery, and 16% will pursue observation. 13 For men whose disease is low risk and it is either unknown if they are metastatic or known that their disease is localized, prior research has documented that 6% will pursue systemic therapy, 52% radiation, 18% surgery, and 23% will receive no treatment/observation ( Figure S1). 14

| Impact of treatment choice on survival
There are some men who do not receive imaging but have metastatic cancer, which is not known at diagnosis. For these men, we assumed that the metastatic disease will be identified within 3 months of diagnosis, and subsequently their treatment approach will change. However, in the time period when the severity of the patient's disease is not known, we assumed that there was an increased risk of death.
We modeled this assuming a hazard rate of two in this three-month period. This was a conservative assumption and biased the results toward the null. We did not assume that imaging would have any other impact on treatment choice or survival.

| Utility of treatments (quality of life)
We used previously published studies to quantify a patient's utility of particular health states 15,16 as shown in Table 1. If patients were in multiple health states simultaneously, such as receiving surgery and having metastatic disease, then we used the minimum method to select the health state with the worst quality of life.

| Costs
Costs assumptions came from multiple studies of prostate cancer (Table 1). [18][19][20] For the costs of specific treatment choices, we assumed that they accrued consistently across the year after diagnosis. For men that changed treatments due to detection of an unknown

| Scenario
We simulated the status quo of widespread inappropriate imaging and compared it to the optimal scenario where only individuals with highrisk prostate cancer received imaging and individual with low-risk prostate cancer did not.  (Table 2). We also found that guidelineconcordant imaging resulted in men with metastatic cancers being much more likely to receive initial systemic therapy (72 vs. 94%) instead of starting on a different therapy and switching to a more appropriate therapy once metastatic disease was identified. This earlier detection of metastases resulted in improved initial treatment and slightly lower first-year treatment costs ($29 993 vs. $29 884, undiscounted). Additionally, we found that improved initial treatment resulted in slightly lower first-year treatment costs stemming from men with low-risk disease and unknown metastasis that would have received imaging. These men would have been diagnosed with metastatic disease under the status quo but would not receive imaging in the ideal scenario. This small group of men has an increased risk of mortality until incidental detection.

| Microsimulation model implementation
The discounted and undiscounted incremental cost-effectiveness ratios indicated that guideline-concordant imaging was cost-saving and slightly more effective compared with current practice patterns.
Moreover, the probabilistic sensitivity analysis showed that 98% of iterations reduced costs. Additionally, we found a very small improvement in undiscounted life years (12.0960 vs. 12.0961 life years). This small difference is observed in the probabilistic sensitivity analysis, which found that only 52% of iterations improved health and the remaining 48% iterations reduced QALYs.
All one-way sensitivity analyses (Figure 1) showed consistent results to the base case. Generally, it appeared that the results were most sensitive to the treatments used for high-risk men and radiation costs. As shown in Figure 2, the probabilistic analysis had results that spanned all quadrants of the incremental cost-effectiveness plane, with 51% of iterations improving health and reducing costs, 47% reducing health and costs, under 1% increasing health and costs, and under 1% decreasing health and increasing costs. The CEAC ( Figure 3) showed that reducing imaging is the preferred strategy at all WTP thresholds; however, the probability of cost-effectiveness declined as the WTP increased. This is largely due to iterations where health and costs were both reduced. As the value of a QALY increased, the savings gained mattered less relative to any health lost.

| DISCUSSION
The incremental cost-effectiveness ratio comparing guidelineconcordant imaging to status quo imaging indicated that guidelineconcordant imaging would be cost saving but would have a limited impact on health. This demonstrates that appropriate reallocation of imaging resources would save the health system money, a situation rarely seen in cancer care. 22 The spending reduction and QALY bene- This underscores the strong need and demand for the present analysis as a novel investigation documenting the cost-benefit trade-offs.
Additionally, in settings with lower rates of inappropriate imaging of low-risk patients, there usually exists similarly lower rates of appropriate imaging for high-risk patients. 10 This suggests that numerous care settings have an opportunity to improve the efficacy and value of the prostate cancer care they provide.
While this model illustrates the excess costs incurred with guideline-discordant care, it does not account for patient preference.
It is possible that imaging could be driven by patient demands in hopes to reduce anxiety; however, a recent qualitative study has found that the decision to pursue imaging is largely driven by physician decision-making. 25 Additionally, research has shown that many men with low-risk disease who receive inappropriate imaging also receive false-positive test results, 3 which has been shown to increase anxiety in other cancers. 26,27 This study demonstrates the potential for cost-savings through appropriate imaging ordering practices. Cost-saving interventions are rare, however. A recent review of cost-effectiveness studies found that only 10% of prostate cancer interventions are cost saving. 22 The cost reduction shown in our model is modest, suggesting that any intervention to improve imaging guideline adherence should be low cost. One potential solution is the Prostate Cancer Imaging Stewardship (PCIS) 28 Intervention, a bundle of evidence-based implementation strategies that target clinician behavior change in order to increase prostate cancer staging imaging guideline concordance.
These findings highlight an opportunity within the healthcare system to reduce unnecessary costs and overtreatment through guideline adherence. In current clinical practice, the imaging recommendations and guidelines remain largely unchanged. There are now newer, more expensive technologies for staging, and it is possible that we will be underestimating the importance and value of guideline-concordant imaging in the field today.