Optimal Donor Selection Across Multiple Outcomes For Hematopoietic Stem Cell Transplantation By Bayesian Nonparametric Machine Learning.

Allogeneic hematopoietic cell transplantation (HCT) is one of the only curative treatment options for patients suffering from life-threatening hematologic malignancies; yet, the possible adverse complications can be serious even fatal. Matching between donor and recipient for 4 of the HLA genes is widely accepted and supported by the literature. However, among 8/8 allele matched unrelated donors, there is less agreement among centers and transplant physicians about how to prioritize donor characteristics like additional HLA loci (DPB1 and DQB1), donor sex/parity, CMV status, and age to optimize transplant outcomes. This leads to varying donor selection practice from patient to patient or via center protocols. Furthermore, different donor characteristics may impact different post transplant outcomes beyond mortality, including disease relapse, graft failure/rejection, and chronic graft-versus-host disease (components of event-free survival, EFS). We develop a general methodology to identify optimal treatment decisions by considering the trade-offs on multiple outcomes modeled using Bayesian nonparametric machine learning. We apply the proposed approach to the problem of donor selection to optimize overall survival and event-free survival, using a large outcomes registry of HCT recipients and their actual and potential donors from the Center for International Blood and Marrow Transplant Research (CIBMTR). Our approach leads to a donor selection strategy that favors the youngest male donor, except when there is a female donor that is substantially younger.

Nonparametric failure time: Time-to-event machine learning with heteroskedastic Bayesian additive regression trees and low information omnibus Dirichlet process mixtures.

Many popular survival models rely on restrictive parametric, or semiparametric, assumptions that could provide erroneous predictions when the effects of covariates are complex. Modern advances in computational hardware have led to an increasing interest in flexible Bayesian nonparametric methods for time-to-event data such as Bayesian additive regression trees (BART). We propose a novel approach that we call nonparametric failure time (NFT) BART in order to increase the flexibility beyond accelerated failure time (AFT) and proportional hazard models. NFT BART has three key features: (1) a BART prior for the mean function of the event time logarithm; (2) a heteroskedastic BART prior to deduce a covariate-dependent variance function; and (3) a flexible nonparametric error distribution using Dirichlet process mixtures (DPM). Our proposed approach widens the scope of hazard shapes including nonproportional hazards, can be scaled up to large sample sizes, naturally provides estimates of uncertainty via the posterior and can be seamlessly employed for variable selection. We provide convenient, user-friendly, computer software that is freely available as a reference implementation. Simulations demonstrate that NFT BART maintains excellent performance for survival prediction especially when AFT assumptions are violated by heteroskedasticity. We illustrate the proposed approach on a study examining predictors for mortality risk in patients undergoing hematopoietic stem cell transplant (HSCT) for blood-borne cancer, where heteroskedasticity and nonproportional hazards are likely present.

Mandibular advancement reduces pharyngeal collapsibility by enlarging the airway rather than affecting velopharyngeal compliance.

Mandibular advancement devices (MADs) are frequently prescribed for obstructive sleep apnea (OSA) patients, but approximately one third of patients experience no therapeutic benefit. Understanding the mechanisms by which MADs prevent pharyngeal collapse may help optimize MAD therapy. This study quantified the relative contributions of changes in airspace cross-sectional area (CSA) versus changes in velopharyngeal compliance in determining MAD efficacy. Sixteen patients with moderate to severe OSA (mean apnea-hypopnea index of 32 ± 15 events/h) underwent measurements of the velopharyngeal closing pressure (PCLOSE ) during drug induced sedated endoscopy (DISE) via stepwise reductions in nasal mask pressure and recording of the intraluminal pressure with a catheter. Airspace CSA was estimated from video endoscopy. Pharyngeal compliance was defined as the slope of the area-pressure relationship of the velopharyngeal airspace. MAD therapy reduced PCLOSE from a median of 0.5 cmH2 O pre-advancement to a median of -2.6 cmH2 O post-advancement (p = 0.0009), increased the minimal CSA at the velopharynx by approximately 20 mm2 (p = 0.0067), but did not have a statistically significant effect on velopharyngeal compliance (p = 0.23). PCLOSE had a strong correlation with CSA but did not correlate with velopharyngeal compliance. Our results suggest that MADs reduce velopharyngeal collapsibility by increasing airway size as opposed to affecting velopharyngeal compliance. This contradicts the speculation of previous literature that the effectiveness of MADs is partially due to a reduction in velopharyngeal compliance resulting from stretching of the soft palate. These findings suggest that quantification of velopharyngeal CSA pre- and post-MAD advancement has potential as a biomarker to predict the success of MAD therapy.

Anatomical determinants of upper airway collapsibility in obstructive sleep apnea: A systematic review and meta-analysis.

Upper airway (UA) collapsibility is one of the key factors that determine the severity of obstructive sleep apnea (OSA). Interventions for OSA are aimed at reducing UA collapsibility, but selecting the optimal alternative intervention for patients who fail CPAP is challenging because currently no validated method predicts how anatomical changes affect UA collapsibility. The gold standard objective measure of UA collapsibility is the pharyngeal critical pressure (Pcrit). A systematic literature review and meta-analysis were performed to identify the anatomical factors with the strongest correlation with Pcrit. A search using the PRISMA methodology was performed on PubMed for English language scientific papers that correlated Pcrit to anatomic variables and OSA severity as measured by the apnea-hypopnea index (AHI). A total of 29 papers that matched eligibility criteria were included in the quantitative synthesis. The meta-analysis suggested that AHI has only a moderate correlation with Pcrit (estimated Pearson correlation coefficient r = 0.46). The meta-analysis identified four key anatomical variables associated with UA collapsibility, namely hyoid position (r = 0.53), tongue volume (r = 0.51), pharyngeal length (r = 0.50), and waist circumference (r = 0.49). In the future, biomechanical models that quantify the relative importance of these anatomical factors in determining UA collapsibility may help identify the optimal intervention for each patient. Many anatomical and structural factors such as airspace cross-sectional areas, epiglottic collapse, and palatal prolapse have inadequate data and require further research.

Novel Pediatric Height Outlier Detection Methodology for Electronic Health Records via Machine Learning With Monotonic Bayesian Additive Regression Trees.

OBJECTIVE: To create a new methodology that has a single simple rule to identify height outliers in the electronic health records (EHR) of children. METHODS: We constructed 2 independent cohorts of children 2 to 8 years old to train and validate a model predicting heights from age, gender, race and weight with monotonic Bayesian additive regression trees. The training cohort consisted of 1376 children where outliers were unknown. The testing cohort consisted of 318 patients that were manually reviewed retrospectively to identify height outliers. RESULTS: The amount of variation explained in height values by our model, R2 , was 82.2% and 75.3% in the training and testing cohorts, respectively. The discriminatory ability to assess height outliers in the testing cohort as assessed by the area under the receiver operating characteristic curve was excellent, 0.841. Based on a relatively aggressive cutoff of 0.075, the outlier sensitivity is 0.713, the specificity 0.793; the positive predictive value 0.615 and the negative predictive value is 0.856. CONCLUSIONS: We have developed a new reliable, largely automated, outlier detection method which is applicable to the identification of height outliers in the pediatric EHR. This methodology can be applied to assess the veracity of height measurements ensuring reliable indices of body proportionality such as body mass index.

Optimal Donor Selection for Hematopoietic Cell Transplantation Using Bayesian Machine Learning.

PURPOSE: Donor selection practices for matched unrelated donor (MUD) hematopoietic cell transplantation (HCT) vary, and the impact of optimizing donor selection in a patient-specific way using modern machine learning (ML) models has not been studied. METHODS: We trained a Bayesian ML model in 10,318 patients who underwent MUD HCT from 1999 to 2014 to provide patient- and donor-specific predictions of clinically severe (grade 3 or 4) acute graft-versus-host disease or death by day 180. The model was validated in 3,501 patients from 2015 to 2016 with archived records of potential donors at search. Donor selection optimizing predicted outcomes was implemented over either an unlimited donor pool or the donors in the search archives. Posterior mean differences in outcomes from optimal donor selection versus actual practice were summarized per patient and across the population with 95% intervals. RESULTS: Event rates were 33% (training) and 37% (validation). Among donor features, only age affected outcomes, with the effect consistent regardless of patient features. The median (interquartile range) difference in age between the youngest donor at search and the selected donor was 6 (1-10) years, whereas the number of donors per patient younger than the selected donor was 6 (1-36). Fourteen percent of the validation data set had an approximate 5% absolute reduction in event rates from selecting the youngest donor at search versus the actual donor used, leading to an absolute population reduction of 1% (95% interval, 0 to 3). CONCLUSION: We confirmed the singular importance of selecting the youngest available MUD, irrespective of patient features, identified potential for improved HCT outcomes by selecting a younger MUD, and demonstrated use of novel ML models transferable to optimize other complex treatment decisions in a patient-specific way.

Non-parametric recurrent events analysis with BART and an application to the hospital admissions of patients with diabetes.

Much of survival analysis is concerned with absorbing events, i.e., subjects can only experience a single event such as mortality. This article is focused on non-absorbing or recurrent events, i.e., subjects are capable of experiencing multiple events. Recurrent events have been studied by many; however, most rely on the restrictive assumptions of linearity and proportionality. We propose a new method for analyzing recurrent events with Bayesian Additive Regression Trees (BART) avoiding such restrictive assumptions. We explore this new method via a motivating example of hospital admissions for diabetes patients and simulated data sets.

The interplay between hospital and surgeon factors and the use of sentinel lymph node biopsy for breast cancer.

BACKGROUND: Several surgeon characteristics are associated with the use of sentinel lymph node biopsy (SLNB) for breast cancer. No studies have systematically examined the relative contribution of both surgeon and hospital factors on receipt of SLNB. OBJECTIVE: To evaluate the relationship between surgeon and hospital characteristics, including a novel claims-based classification of hospital commitment to cancer care (HC), and receipt of SLNB for breast cancer, a marker of quality care. DATA SOURCES/STUDY DESIGN: Observational prospective survey study was performed in a population-based cohort of Medicare beneficiaries who underwent incident invasive breast cancer surgery, linked to Medicare claims, state tumor registries, American Hospital Association Annual Survey Database, and American Medical Association Physician Masterfile. Multiple logistic regression models determined surgeon and hospital characteristics that were predictors of SLNB. RESULTS: Of the 1703 women treated at 471 different hospitals by 947 different surgeons, 65% underwent an initial SLNB. Eleven percent of hospitals were high-volume and 58% had a high commitment to cancer care. In separate adjusted models, both high HC (odds ratio [OR] 1.53, 95% confidence interval [CI] 1.12-2.10) and high hospital volume (HV, OR 1.90, 95% CI 1.28-2.79) were associated with SLNB. Adding surgeon factors to a model including both HV and HC minimally modified the effect of high HC (OR 1.34, 95% CI 0.95-1.88) but significantly weakened the effect of high HV (OR 1.25, 95% CI 0.82-1.90). Surgeon characteristics (higher volume and percentage of breast cancer cases) remained strong independent predictors of SLNB, even when controlling for various hospital characteristics. CONCLUSIONS: Hospital factors are associated with receipt of SLNB but surgeon factors have a stronger association. Since regionalization of breast cancer care in the U.S. is unlikely to occur, efforts to improve the surgical care and outcomes of breast cancer patients must focus on optimizing patient access to SLNB by ensuring hospitals have the necessary resources and training to perform SLNB, staffing hospitals with surgeons who specialize/focus in breast cancer and referring patients who do not have access to SLNB to an experienced center.

Nonparametric survival analysis using Bayesian Additive Regression Trees (BART).

Bayesian additive regression trees (BART) provide a framework for flexible nonparametric modeling of relationships of covariates to outcomes. Recently, BART models have been shown to provide excellent predictive performance, for both continuous and binary outcomes, and exceeding that of its competitors. Software is also readily available for such outcomes. In this article, we introduce modeling that extends the usefulness of BART in medical applications by addressing needs arising in survival analysis. Simulation studies of one-sample and two-sample scenarios, in comparison with long-standing traditional methods, establish face validity of the new approach. We then demonstrate the model's ability to accommodate data from complex regression models with a simulation study of a nonproportional hazards scenario with crossing survival functions and survival function estimation in a scenario where hazards are multiplicatively modified by a highly nonlinear function of the covariates. Using data from a recently published study of patients undergoing hematopoietic stem cell transplantation, we illustrate the use and some advantages of the proposed method in medical investigations. Copyright © 2016 John Wiley & Sons, Ltd.

An algorithm to identify the development of lymphedema after breast cancer treatment.

PURPOSE: Large, population-based studies are needed to better understand lymphedema, a major source of morbidity among breast cancer survivors. One challenge is identifying lymphedema in a consistent fashion. We sought to develop and validate an algorithm using Medicare claims to identify lymphedema after breast cancer surgery. METHODS: From a population-based cohort of 2,597 elderly (65+) women who underwent incident breast cancer surgery in 2003 and completed annual telephone surveys through 2008, two algorithms were developed using Medicare claims from half of the cohort and validated in the remaining half. A lymphedema-positive case was defined by patient report. RESULTS: A simple two ICD-9 code algorithm had 69 % sensitivity, 96 % specificity, positive predictive value >75 % if prevalence of lymphedema is >16 %, negative predictive value >90 %, and area under receiver operating characteristic curve (AUC) of 0.82 (95 % CI 0.80-0.85). A more sophisticated, multi-step algorithm utilizing diagnostic and treatment codes, logistic regression methods, and a reclassification step performed similarly to the two-code algorithm. CONCLUSIONS: Given the similar performance of the two validated algorithms, the ease of implementing the simple algorithm and the fact that the simple algorithm does not include treatment codes, we recommend that this two-code algorithm be validated in and applied to other population-based breast cancer cohorts. IMPLICATIONS FOR CANCER SURVIVORS: This validated lymphedema algorithm will facilitate the conduct of large, population-based studies in key areas (incidence rates, risk factors, prevention measures, treatment, and cost/economic analyses) that are critical to advancing our understanding and management of this challenging and debilitating chronic disease.

Surgeon specialization and use of sentinel lymph node biopsy for breast cancer.

IMPORTANCE: Sentinel lymph node biopsy (SLNB) is the standard of care for axillary staging in patients with clinically node-negative breast cancer. It is not known whether SLNB rates differ by surgeon expertise. If surgeons with less breast cancer expertise are less likely to offer SLNB to these patients, this practice pattern could lead to unnecessary axillary lymph node dissections and lymphedema. OBJECTIVE: To explore potential measures of surgical expertise (including a novel objective specialization measure: percentage of a surgeon's operations performed for breast cancer determined from Medicare claims) on the use of SLNB for invasive breast cancer. DESIGN, SETTING, AND POPULATION: A population-based prospective cohort study was conducted in California, Florida, and Illinois. Participants included elderly (65-89 years) women identified from Medicare claims as having had incident invasive breast cancer surgery in 2003. Patient, tumor, treatment, and surgeon characteristics were examined. MAIN OUTCOME AND MEASURE: Type of axillary surgery performed. RESULTS: Of 1703 women who received treatment by 863 surgeons, 56.4% underwent an initial SLNB, 37.2% initial axillary lymph node dissection, and 6.3% no axillary surgery. The median annual surgeon Medicare volume of breast cancer cases was 6.0 (range, 1.5-57.0); the median surgeon percentage of breast cancer cases was 4.5% (range, 0.4%-100.0%). After multivariable adjustment of patient and surgeon factors, women operated on by surgeons with higher volumes and percentages of breast cancer cases had a higher likelihood of undergoing SLNB. Specifically, women were most likely to undergo SLNB if the operation was performed by high-volume surgeons (regardless of percentage) or by lower-volume surgeons with a high percentage of breast cancer cases. In addition, membership in the American Society of Breast Surgeons (odds ratio, 1.98; 95% CI, 1.51-2.60) and Society of Surgical Oncology (1.59; 1.09-2.30) were independent predictors of women undergoing an initial SLNB. CONCLUSIONS AND RELEVANCE: Patients who receive treatment from surgeons with more experience with and focus on breast cancer are significantly more likely to undergo SLNB, highlighting the importance of receiving initial treatment by specialized providers. Factors relating to specialization in a particular area, including our novel surgeon percentage measure, require further investigation as potential indicators of quality of care.

Thirty-day readmissions after elective spine surgery for degenerative conditions among US Medicare beneficiaries.

BACKGROUND CONTEXT: Readmissions within 30 days of hospital discharge are undesirable and costly. Little is known about reasons for and predictors of readmissions after elective spine surgery to help plan preventative strategies. PURPOSE: To examine readmissions within 30 days of hospital discharge, reasons for readmission, and predictors of readmission among patients undergoing elective cervical and lumbar spine surgery for degenerative conditions. STUDY DESIGN: Retrospective cohort study. PATIENT SAMPLE: Patient sample includes 343,068 Medicare beneficiaries who underwent cervical and lumbar spine surgery for degenerative conditions from 2003 to 2007. OUTCOME MEASURES: Readmissions within 30 days of discharge, excluding readmissions for rehabilitation. METHODS: Patients were identified in Medicare claims data using validated algorithms. Reasons for readmission were classified into clinically meaningful categories using a standardized coding system (Clinical Classification Software). RESULTS: Thirty-day readmissions were 7.9% after cervical surgery and 7.3% after lumbar surgery. There was no dominant reason for readmissions. The most common reasons for readmissions were complications of surgery (26%-33%) and musculoskeletal conditions in the same area of the operation (15%). Significant predictors of readmission for both operations included older age, greater comorbidity, dual eligibility for Medicare/Medicaid, and greater number of fused levels. For cervical spine readmissions, additional risk factors were male sex, a diagnosis of myelopathy, and a posterior or combined anterior/posterior surgical approach; for lumbar spine readmissions, additional risk factors were black race, Middle Atlantic geographic region, fusion surgery, and an anterior surgical approach. Our model explained more than 60% of the variability in readmissions. CONCLUSIONS: Among Medicare beneficiaries, 30-day readmissions after elective spine surgery for degenerative conditions represent a target for improvement. Both patient factors and operative techniques are associated with readmissions. Interventions to minimize readmissions should be specific to surgical site and focus on high-risk subgroups where clinical trials of interventions may be of greatest benefit.

Socioeconomic and racial differences in treatment for breast cancer at a low-volume hospital.

PURPOSE: Population-based studies have revealed higher mortality among breast cancer patients treated in low-volume hospitals. Other studies have demonstrated disparities in race and socioeconomic status (SES) in breast cancer survival. The purpose of our study was to determine whether nonwhite or low-SES patients are disproportionately treated in low-volume hospitals. METHODS: A population-based cohort of 2,777 Medicare breast cancer patients who underwent breast cancer surgery in 2003 participated in a survey study examining breast cancer outcomes. Information was obtained from survey responses, Medicare claims, and state tumor registry data. RESULTS: On univariate analysis, patients treated at low-volume hospitals were less likely to be white, less likely to live in an urban location, and more likely to have a low SES with less social support and live a greater distance from a high-volume hospital. Education, marital status, total household income, having additional insurance besides Medicare, population density of primary residence, and tangible support were associated with distance to the nearest high-volume hospital. On multivariate analysis, the independent predictors of treatment at a low-volume hospital were being nonwhite (P = 0.003), having a lower household income (P < 0.0001), residence in a rural location (P = 0.01), and living a greater distance from a high-volume hospital (P < 0.0001). CONCLUSIONS: In this large population-based cohort, women who were poorer, nonwhite, and who lived in a rural location or at a greater distance from a high-volume hospital were more likely to be treated at low-volume hospitals. These differences may partially explain racial and SES disparities in breast cancer outcomes.

Socioeconomic factors associated with adjuvant hormone therapy use in older breast cancer survivors.

BACKGROUND: The authors sought to identify socioeconomic (SES) factors associated with adjuvant hormone therapy (HT) use among a contemporary population of older breast cancer survivors. METHODS: Telephone surveys were conducted among women (ages 65-89 years) residing in 4 states (California, Florida, Illinois, and New York) who underwent initial breast cancer surgery in 2003. Demographic, SES, and treatment information was collected. RESULTS: Of 2191 women, 67% received adjuvant HT with either tamoxifen or an aromatase inhibitor (AI); 71% of those women were on an AI. When adjusting for multiple demographic and SES factors, predictors of HT use were better education (high school degree or higher), better informational/emotional support, and younger age (ages 65-79 years). Race/ethnicity, income, and insurance coverage for medication costs were not associated with receiving HT. For those on HT, when adjusting for all other factors, women were more likely to receive an AI if they had insurance coverage for some or all medication costs, if they were wealthier, if they had better informational/emotional support, and if they were younger (ages 65-69 years). CONCLUSIONS: The majority of older women in this population-based cohort received adjuvant HT, and the adoption of AIs was early. The results indicted that providers should be aware that a woman's education level and support system influence her decision to take HT. Given the high cost of AIs, their benefits in postmenopausal women with hormone receptor-positive breast cancer, and the current finding that women with no insurance coverage for medication costs were significantly less likely to receive an AI, we recommend that policymakers address this issue.

Heightened attention to medical privacy: challenges for unbiased sample recruitment and a possible solution.

Subject recruitment for epidemiologic studies is associated with major challenges due to privacy laws now common in many countries. Privacy policies regarding recruitment methods vary tremendously across institutions, partly because of a paucity of information about what methods are acceptable to potential subjects. The authors report the utility of an opt-out method without prior physician notification for recruiting community-dwelling US women aged 65 years or older with incident breast cancer in 2003. Participants (n = 3,083) and possibly eligible nonparticipants (n = 2,664) were compared using characteristics derived from billing claims. Participation for persons with traceable contact information was 70% initially (2005-2006) and remained over 90% for 3 follow-up surveys (2006-2008). Older subjects and those living in New York State were less likely to participate, but participation did not differ on the basis of socioeconomic status, race/ethnicity, underlying health, or type of cancer treatment. Few privacy concerns were raised by potential subjects, and no complaints were lodged. Using opt-out methods without prior physician notification, a population-based cohort of older breast cancer subjects was successfully recruited. This strategy may be applicable to population-based studies of other diseases and is relevant to privacy boards making decisions about recruitment strategies acceptable to the public.

Exploring the surgeon volume outcome relationship among women with breast cancer.

BACKGROUND: A relationship between higher surgeon volume and lower mortality has been described for breast cancer, but selection bias has not been rigorously evaluated. We studied potential bias in the surgeon volume-outcome relationship by comparing the relationship of surgeon volume to breast cancer mortality and to mortality from other causes of death. METHODS: We conducted an observational cohort study from tumor registry and Medicare claims data on 12 216 women, 66 years or older, with stage I or II breast cancer, who were operated on by 1856 surgeons. Breast cancer mortality and other-cause mortality were determined from death certificate sources and surgeon volume from Medicare claims. RESULTS: Treatment by a high-volume surgeon was associated with younger patient age, white race, less comorbidity, and residence in a more affluent zip code. Patients treated by low-, medium-, and high-volume surgeons had small differences in breast cancer mortality (17.4, 15.7, and 13.0 deaths per 1000 person-years, respectively; P = .03) but larger differences in non-breast cancer mortality (46.0, 36.8, and 31.7 deaths per 1000 person-years, respectively; P < .001). After adjustment for multiple patient and disease factors, women treated by high-volume surgeons, compared with those treated by low-volume surgeons, were not less likely to die of breast cancer (relative risk, 0.94; 95% confidence interval, 0.76-1.16) but were significantly less likely to die of other causes (relative risk, 0.86; 95% confidence interval, 0.75-0.98). CONCLUSIONS: The surgeon volume-outcome relationship for these patients with breast cancer was attributable not to mortality from breast cancer but to other causes of death. The lack of specificity of this relationship raises the possibility of selection bias as an explanatory factor.

Disparities in colon cancer screening in the Medicare population.

BACKGROUND: Colorectal cancer is the third most common cancer in the United States, but the rate of screening remains low. Since 2001, Medicare has provided coverage of colonoscopy for colorectal cancer screening in individuals at average risk, but little is known about the effect of this coverage on screening or disparities in screening practices. METHODS: We examined the Medicare physician/supplier billing claims file for New York, Florida, and Illinois for the years 2002 and 2003. Using a previously employed algorithm, we identified the rates of colorectal screening tests in individuals at average risk. We performed multivariate logistic regression analysis to calculate the effects of sex, racial/ethnic, and socioeconomic characteristics on screening. We also looked for interactions between socioeconomic and demographic variables. RESULTS: A total of 596 470 Medicare beneficiaries were included in the study. Approximately 18.3% of the population had undergone a screening colon test during the study period. Nonwhite persons were less likely to be screened for colorectal cancer than were white persons (relative risk [RR], 0.52; 95% confidence interval [CI], 0.50-0.53). The lowest RR of screening colonoscopy in women compared with men was in the oldest age group and the highest income tertile (RR for whites, 0.64; 95% CI, 0.59-0.70). Higher income level was associated with screening colonoscopy in white patients (men: RR, 1.19; 95% CI, 1.14-1.25; women: RR, 1.09; 95% CI, 1.05-1.15) but not in nonwhite patients (men: RR, 0.97; 95% CI, 0.78-1.22; women: RR, 0.94; 95% CI, 0.78-1.14). CONCLUSION: Despite the expansion of Medicare coverage for colorectal cancer screening, there still remain significant disparities between sex and racial/ethnic groups in screening practices.

Bone density testing in older women and its association with patient age.

OBJECTIVES: To measure the early adoption of bone density testing and examine the association between older age and such testing. DESIGN: Retrospective study of Medicare administrative claims. SETTING: Five states and six urban regions in the United States. PARTICIPANTS: Female Medicare recipients aged 66 to 90. MEASUREMENTS: Bone density testing in women without prior osteoporosis or fracture (osteoporosis screening) was evaluated. The association between age and osteoporosis screening was then examined while controlling for other demographic and health factors. RESULTS: Of 43,802 women eligible for osteoporosis screening, 22.9% were tested between 1999 and 2001, the first 3 full years of Medicare coverage. Receipt of bone density tests decreased with increasing age, from 27.2% of women aged 66 to 70 to fewer than 10% of women aged 86 to 90. After adjustment for race, comorbidity, fracture risk, and socioeconomic factors, bone density testing decreased significantly with each age category, so that women aged 71 to 75 were slightly less likely than (adjusted odds ratio (AOR) =0.91, 95% confidence interval (CI) =0.86-0.96), women aged 76 to 80 were less likely than (AOR =0.71, 95% CI =0.67-0.76), and women aged 81 to 85 were half as likely as (AOR =0.50, 95% CI =0.46-0.55) women aged 66 to 70 to receive a bone density test. CONCLUSION: In the 3 years after Medicare reimbursement for osteoporosis screening began, adoption of bone density testing was lowest in women in age groups at highest fracture risk.

An algorithm for the use of Medicare claims data to identify women with incident breast cancer.

OBJECTIVE: To develop and validate a clinically informed algorithm that uses solely Medicare claims to identify, with a high positive predictive value, incident breast cancer cases. DATA SOURCE: Population-based Surveillance, Epidemiology, and End Results (SEER) Tumor Registry data linked to Medicare claims, and Medicare claims from a 5 percent random sample of beneficiaries in SEER areas. STUDY DESIGN: An algorithm was developed using claims from 1995 breast cancer patients from the SEER-Medicare database, as well as 1995 claims from Medicare control subjects. The algorithm was validated on claims from breast cancer subjects and controls from 1994. The algorithm development process used both clinical insight and logistic regression methods. DATA EXTRACTION: Training set: Claims from 7,700 SEER-Medicare breast cancer subjects diagnosed in 1995, and 124,884 controls. Validation set: Claims from 7,607 SEER-Medicare breast cancer subjects diagnosed in 1994, and 120,317 controls. PRINCIPAL FINDINGS: A four-step prediction algorithm was developed and validated. It has a positive predictive value of 89 to 93 percent, and a sensitivity of 80 percent for identifying incident breast cancer. The sensitivity is 82-87 percent for stage I or II, and lower for other stages. The sensitivity is 82-83 percent for women who underwent either breast-conserving surgery or mastectomy, and is similar across geographic sites. A cohort identified with this algorithm will have 89-93 percent incident breast cancer cases, 1.5-6 percent cancer-free cases, and 4-5 percent prevalent breast cancer cases. CONCLUSIONS: This algorithm has better performance characteristics than previously proposed algorithms. The ability to examine national patterns of breast cancer care using Medicare claims data would open new avenues for the assessment of quality of care.