ORAL ABSTRACTS: CHD, CVD AND CANCER
Method(s): We use ICD-9-CM diagnosis codes to identify prostate cancer patients receiving care at an academic medical center. Patients are confirmed in the California Cancer Registry, which returns tumor characteristic and treatment data on all patients with a confirmed cancer diagnosis, including curated pathology and tumor staging information. Using all proposed prostate cancer quality metrics, we define each quality metric using target terms and concepts to extract from the EHRs. These terms may include diagnostic procedures and tests and their results (such as Digital Rectal Exam, DRE), therapeutic procedures, and drugs (both ordered and administered). Terms are mapped to a standardized medical vocabulary, enabling us to represent the elements of a metric by a concept domain and its permissible values. The structured representation of the quality metric data elements are used to create quality phenotypes, which are rules involving the temporal order of components of the quality metrics.
Result(s): We have developed an EHR database that draws healthcare records from an academic center and link these records to the California Cancer Registry. This allows for clinical care data to be analyzed alongside diagnostic details, which are not usually captured in EHR. This database includes unstructured clinician notes to ensure the broad evaluation of patient-centered data. Furthermore, our system enables real-time extraction of treatment processes and outcome measures, allowing us to use EHR data to track process improvements.
The quality metric phenotypes we create are standardized code that can be used across different EHR systems. For example, the algorithm to detect DRE documentation contains prostate cancer diagnosis code (ICD-9 185), dates (ensure the DRE was performed prior to treatment), and textual concepts (e.g. DRE, digital rectal exam, and rectal exam).
EHR-systems can be used to assess and report quality metrics systematically, efficiently, and with high accuracy. The development of such systems moves the quality assessment field into large-scale analyses.
To project the long-term effectiveness of intensive imaging-surveillance after curative resection of colorectal cancer (CRC). Intensive imaging surveillance consists of annual abdominal (abdominal-pelvic for rectal cancer) and thoracic CT exam for five years post initial curative resection.
We developed a state-transition model that simulated newly diagnosed patients with CRC through a series of disease states characterized by (1) a disease free state, (2) the presence of occult micro-metastasis (undetectable by CT), (3) resectable preclinical macro-metastasis (asymptomatic mets potentially detectable by CT and if found is treatable), (4) unresectable pre-clinical macro-metastasis (asymptomatic mets potentially detectable by CT and if found not amenable to curative treatment), (5) symptomatic and resectable disease, (6) symptomatic and unresectable disease, and (7) death (from cancer and other causes). The microsimulation model utilized nonparametric hazard functions to flexibly model the key underlying time to event processes. These were calibrated simultaneously to Surveillance, Epidemiology, and End Results Program (SEER) stage-specific relative survival and the efficacy of imaging-based surveillance taken from a meta-analysis of multiple randomized clinical trials (RCTs). SEER provides relative survival information, which reflects the underlying cancer mortality rates over time. There have been 11 RCTs of intensive surveillance following CRC diagnosis, 8 of which included intensive imaging follow-up (CT, ultrasound, or chest x-ray). The purpose of conducting the CT is to diagnose metastatic recurrence before it becomes symptomatic when it has a greater chance of being resectable (and thereby reducing cancer mortality).
Calibrated parameter values showed that the median time to detectable mets was 7 months and the median time to symptomatic disease was 13 months, among those patients destined to develop mets. The model calibrated reasonably well to SEER 5-year relative survival for stage III. A strategy of performing annual CT for five years after CRC diagnosis and curative resection resulted in life expectancy gains of about 9 months for stage III rectal cancer and 12 months for stage III colon cancer. For stage II disease, the absolute improvement in LE was reduced by a factor of about 2, with a LE gain of 4 months for rectal cancer and 5 months for colon cancer.
Intensive imaging follow-up after curative resection for stage II-III colorectal cancer can effectively reduce cancer-related mortality and thereby increase life-expectancy.
Purpose: Approximately 25% of rectal cancer patients with a clinical complete response (CR) after neoadjuvant chemoradiation will ultimately be found to have a “true” pathologic CR at both the primary tumor site and the mesorectum at the time of surgical resection (SR; which requires partial/total removal of the rectum). Minimally invasive, transanal local excision (LE) of the primary tumor site after chemoradiation may identify patients who achieved a true pathologic CR after neoadjuvant therapy, and thus, may not require (and/or benefit from) SR. The purpose of this study was to explore the predictive/therapeutic value of selective SR (i.e., based on results of LE) versus routine SR in this patient population.
Method(s): We developed a decision analysis/Markov model to compare outcomes following selective versus SR in patients with mid-low rectal cancers with a clinical CR after chemoradiation. All patients in the selective SR strategy underwent LE after chemoradiation: patients with a pathologic CR at the primary tumor site were observed, while those with residual disease at the primary tumor site underwent subsequent SR. All patients in the routine SR strategy underwent upfront resection after chemoradiation. Sensitivity/specificity of LE, morbidity/mortality of LE/SR, local/systemic recurrence estimates after LE/SR, rates of surgical salvage after local recurrence following LE/SR, and survival estimates were obtained from the medical literature. Model outcomes were quality-adjusted using health state preferences.
Result(s): Overall, unadjusted and quality-adjusted life expectancy was superior after selective SR compared to routine SR; patients with a true pathologic CR gained the greatest benefit (Table 1). Selective SR was the optimal strategy even after model estimates/utilities were varied widely over their reported ranges. Routine SR was preferred only if model estimates/utilities were varied well beyond their reported ranges: if mortality of LE, probability of a true pathologic CR, and utility of being disease-free after LE (without subsequent SR) were assumed to be >4.5%, <0.2%, and 0.846, respectively.
Conclusion(s): Selective SR (based on results of LE) maximizes unadjusted and quality-adjusted life expectancy compared to upfront routine SR in patients with mid-low rectal cancers with a clinical CR after neoadjuvant chemoradiation. Routine SR in this increasingly common clinical situation should be reconsidered. Randomized trials comparing selective versus routine SR (that prospectively measure resulting health-state preferences and costs) in this setting are warranted.
Purpose: Current approaches to defining treatment or diagnostic thresholds are commonly based on average effects, which may lead to incorrect decisions on individual level. We demonstrate a general approach to identify treatment or diagnostic thresholds optimizing individual health outcomes, illustrated for statin treatment based on 10-year coronary heart disease (CHD) risk predicted by the Framingham risk score (FRS).
Method(s): A health economic model was created to evaluate risk-based preventive statin treatment. Based on the Atherosclerosis Risk in Communities study cohorts of men and women aged 50–59 years at low-intermediate or high CHD-risk were simulated and followed for 30 years. Strategies gradually including more individuals by lowering the treatment threshold T (20%-0%;1% decrements) were compared. Differences in health outcomes, quality-adjusted life-years (QALYs) and cost-effectiveness, were assessed at each step to identify optimal treatment thresholds. Cost-effectiveness was evaluated by calculating the net health benefit (NHB) for a willingness-to-pay of $50,000/QALY. At every threshold T both incremental (compared to T=20%) and marginal (compared to T=T+1%) outcomes were evaluated.
Result(s): QALYs ranged from 12.621 in men and 13.696 in women at T=20% to a maximum of 12.689 in men at T=1% and 13.734 in women at T=0%. Keeping the population-level fraction of statin-induced complications <10% resulted in thresholds of T=6% for men and T=2% for women. Lowering the threshold and comparing outcomes after each 1% decrease, QALYs were gained down to T=1% for men and T=0% for women. The incremental NHB was favorable for every threshold down to T=0% among men and down to T=2% among women (Figure 1A). The incremental NHB achieved a maximum at T=3% for men and at T=6% for women, with a NHB of 3,919 and 834 QALYs among 100,000 men and women, respectively. Correspondingly, the marginal NHB was favorable down to T=3% for men and T=6% for women (Figure 1B).
Conclusion(s): Many approaches can be taken to arrive at a treatment or diagnostic threshold. However, current intuition-based approaches leave ample room for health gain and cost savings. Using a stepwise risk-based approach to threshold optimization allows for treatment and diagnostic strategies that optimize outcomes in all individuals instead of on average. This approach can be applied to any outcome, such as to limit complications or missed diagnoses, to maximize health outcomes, or to optimize cost-effectiveness.
Method(s): Data from 20.423 participants of the MORGEN cohort was used and classified into subgroups based on age and gender. CVD events were identified for four prediction models: ATP, Framingham, PCE and SCORE. The 10-year CVD risks and associated burdens, expressed as Quality-Adjusted Life Years (QALYs) lost, were determined and presented for high-risk individuals, i.e. the 25% individuals with highest predicted risks. The effect of a hypothetical (risk factor) treatment in high-risk individuals was investigated, regarding an overall and event specific risk reduction.
Result(s): The distribution of CVD event types varied between men and women but not with age. The predicted risks, as expected, differed substantially with gender and age. Consequently, the predicted burden varied between men and women, and between age-groups, mainly due to differences in predicted risks. For high-risk individuals, men each lost 0.22, 0.83, 0.18, and 0.33 QALYs according to ATP, Framingham, PCE, and SCORE, and women lost 0.07, 0.50, 0.17, 0.13 QALYs, respectively. When treating these high-risk individuals, the burden for men decreased to 0.14, 0.54, 0.12, and 0.22 QALYs lost, and for women to 0.05, 0.33, 0.11, and 0.09 QALYs lost, according to ATP, Framingham, PCE, and SCORE, respectively.
Conclusion(s): Estimates of CVD burden depend as much on the CVD event types included in risk prediction models as on the risk estimates produced by such models. Investigating the distribution of CVD events occurring in practice is therefore necessary to obtain robust estimates of CVD burden and the potential reduction from preventive strategies. Furthermore, as the risks and consequences of specific CVD events are demonstrated to differ for gender and age, evidence of the distribution of CVD events should be obtained for the considered population targeted for preventive strategies.
Purpose: We demonstrate an approach to assess the impact of uncertainty in risk predictions on health-economic outcomes in risk-stratified prevention strategies, illustrated for preventive statin treatment based on 10-year coronary heart disease (CHD) risk predicted by the Framingham risk score (FRS).
Method(s): A Markov decision-analytic model was used to simulate cohorts with preventive statin treatment. We fitted the FRS to men and women from the Atherosclerosis Risk in Communities (ARIC) cohort. Using the ARIC risk distributions hypothetical cohorts of men and women aged 50–59 years followed for 30 years. Individuals were preventively treated if their predicted CHD risk exceeded treatment threshold T. While lowering the threshold T from 20% to 0% (1% decrements), strategies including gradually more and more treated individuals were evaluated. Assessing quality-adjusted life-years (QALYs) and costs at each step, the Net Health Benefit (NHB) (willingness-to-pay of $50,000/QALY) of treating an individual with a certain predicted risk was calculated.
Subsequently, the FRS was refitted to 1,000 bootstrap samples of men and women from the ARIC cohort of varying sizes (N=5,000;N=2,500;N=1,000), while rejecting models not achieving en acceptable level of performance (AUC>0.6). Using these refitted models, we calculated 1,000 alternative individual risk predictions. We then assessed whether a different treatment decision would have been made when applying the alternative risk predictions. Finally, we matched each alternative risk prediction to the corresponding NHB to estimate the impact of the uncertainty a predicted risk on the NHB.
Result(s): Preliminary results indicate that prediction uncertainty resulted in probabilities of incorrect treatment decisions of up to 0.34 and 0.47 (N=5,000), 0.40 and 0.49 (N=2,500), and 0.47 and 0.55 (N=1,000) for predicted risks surrounding T=5% and T=20%, respectively (Figure). The risk-based NHBs ranged from K for a predicted risk p=L% to K for p=L% in men and from K at p=L% to K for p=L% in women.
Conclusion(s): While uncertainty in risk predictions may lead to incorrect treatment decisions, associated impact on long-term health-economic outcomes is often unknown. Assessing this impact can guide studies aiming to improve prediction models by focusing on improving risk prediction in individuals for which improvement may actually improve health-economic outcomes.