Current Role at Stanford
Education & Certifications
MPH, Boston University, Epidemiology/ Biostatistics (2000)
Ph.D, Boston University, Biostatistics (2009)
Mathematical Statistician, NIH/ NCI/ DCCPS/ SRP (6/2011 - 7/2015)
Program Director and Biostatistician for the SEER program.
Rockville, MD USA
Post Doctoral Research Fellow, Harvard University Medical School (2010 - 2011)
Awarded T32 to study competing risk in Total Hip Replacement
Biostatistician/ Programmer, Brigham & Women's Hospital (2005 - 2010)
Programmer and Biostatistician for Division of Pharmacoepidemiology and Pharmacoeconomics at Brigham and Women's Hospital
Professional Affiliations and Activities
member, American Statistical Association (1998 - Present)
Current estimates of the cure fraction: a feasibility study of statistical cure for breast and colorectal cancer.
Journal of the National Cancer Institute. Monographs
2014; 2014 (49): 244-254
The probability of cure is a long-term prognostic measure of cancer survival. Estimates of the cure fraction, the proportion of patients "cured" of the disease, are based on extrapolating survival models beyond the range of data. The objective of this work is to evaluate the sensitivity of cure fraction estimates to model choice and study design.Data were obtained from the Surveillance, Epidemiology, and End Results (SEER)-9 registries to construct a cohort of breast and colorectal cancer patients diagnosed from 1975 to 1985. In a sensitivity analysis, cure fraction estimates are compared from different study designs with short- and long-term follow-up. Methods tested include: cause-specific and relative survival, parametric mixture, and flexible models. In a separate analysis, estimates are projected for 2008 diagnoses using study designs including the full cohort (1975-2008 diagnoses) and restricted to recent diagnoses (1998-2008) with follow-up to 2009.We show that flexible models often provide higher estimates of the cure fraction compared to parametric mixture models. Log normal models generate lower estimates than Weibull parametric models. In general, 12 years is enough follow-up time to estimate the cure fraction for regional and distant stage colorectal cancer but not for breast cancer. 2008 colorectal cure projections show a 15% increase in the cure fraction since 1985.Estimates of the cure fraction are model and study design dependent. It is best to compare results from multiple models and examine model fit to determine the reliability of the estimate. Early-stage cancers are sensitive to survival type and follow-up time because of their longer survival. More flexible models are susceptible to slight fluctuations in the shape of the survival curve which can influence the stability of the estimate; however, stability may be improved by lengthening follow-up and restricting the cohort to reduce heterogeneity in the data.
View details for DOI 10.1093/jncimonographs/lgu015
View details for PubMedID 25417238
Adolescent and young adult cancer survival.
Journal of the National Cancer Institute. Monographs
2014; 2014 (49): 228-235
Adolescent and young adults (AYAs) face challenges in having their cancers recognized, diagnosed, treated, and monitored. Monitoring AYA cancer survival is of interest because of the lack of improvement in outcome previously documented for these patients as compared with younger and older patient outcomes. AYA patients 15-39 years old, diagnosed during 2000-2008 with malignant cancers were selected from the SEER 17 registries data. Selected cancers were analyzed for incidence and five-year relative survival by histology, stage, and receptor subtypes. Hazard ratios were estimated for cancer death risk among younger and older ages relative to the AYA group. AYA survival was worse for female breast cancer (regardless of estrogen receptor status), acute lymphoid leukemia (ALL), and acute myeloid leukemia (AML). AYA survival for AML was lowest for a subtype associated with a mutation of the nucleophosmin 1 gene (NPM1). AYA survival for breast cancer and leukemia remain poor as compared with younger and older survivors. Research is needed to address disparities and improve survival in this age group.
View details for DOI 10.1093/jncimonographs/lgu019
View details for PubMedID 25417236
A comparison of statistical approaches for physician-randomized trials with survival outcomes
CONTEMPORARY CLINICAL TRIALS
2012; 33 (1): 104-115
This study compares methods for analyzing correlated survival data from physician-randomized trials of health care quality improvement interventions. Several proposed methods adjust for correlated survival data; however the most suitable method is unknown. Applying the characteristics of our study example, we performed three simulation studies to compare conditional, marginal, and non-parametric methods for analyzing clustered survival data. We simulated 1000 datasets using a shared frailty model with (1) fixed cluster size, (2) variable cluster size, and (3) non-lognormal random effects. Methods of analyses included: the nonlinear mixed model (conditional), the marginal proportional hazards model with robust standard errors, the clustered logrank test, and the clustered permutation test (non-parametric). For each method considered we estimated Type I error, power, mean squared error, and the coverage probability of the treatment effect estimator. We observed underestimated Type I error for the clustered logrank test. The marginal proportional hazards method performed well even when model assumptions were violated. Nonlinear mixed models were only advantageous when the distribution was correctly specified.
View details for DOI 10.1016/j.cct.2011.08.008
View details for Web of Science ID 000300072500018
View details for PubMedID 21924382
- Meta-analyses involving cross-over trials: methodological issues INTERNATIONAL JOURNAL OF EPIDEMIOLOGY 2011; 40 (6): 1732-1734
A SAS macro for a clustered logrank test
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
2011; 104 (2): 266-270
The clustered logrank test is a nonparametric method of significance testing for correlated survival data. Examples of its application include cluster randomized trials where groups of patients rather than individuals are randomized to either a treatment or a control intervention. We describe a SAS macro that implements the 2-sample clustered logrank test for data where the entire cluster is randomized to the same treatment group. We discuss the theory and applications behind this test as well as details of the SAS code.
View details for DOI 10.1016/j.cmpb.2011.02.001
View details for Web of Science ID 000296945100031
View details for PubMedID 21496938
Guideline-conformity of initiation with oral hypoglycemic treatment for patients with newly therapy-dependent type 2 diabetes mellitus in Austria
PHARMACOEPIDEMIOLOGY AND DRUG SAFETY
2011; 20 (1): 57-65
To determine guideline conformity of initiation of oral hypoglycemic (OH) treatment for type 2 diabetes in Austria; to study patient and prescriber correlates of recommended initiation with metformin monotherapy.We used claims from 11 sickness funds that covered 7.5 million individuals, representing >90% of the Austrian population. First-time OH use was defined as a first filled prescription after one year without any OH drug or insulin. Among these incident users, we described the OH drug class used and defined correlates of initiation with metformin monotherapy.From 1/2007 to 6/2008, we identified 42,882 incident users of an OH drug: 70.8% used metformin, 24.7% used a sulfonylurea, and 4.5% initiated treatment with another class. We estimated the incidence of OH-dependent type 2 diabetes at 3.8-4.4 per 1000 patient-years. We conducted multivariate analyses among 39 077 patients with available prescriber information. Independent correlates of initiation with metformin were younger age, female gender, waived co-payment, more recent initiation, fewer hospital days and more therapeutic classes received in the year prior to first OH therapy (all p < 0.001). Prescriber specialty and age (p < 0.001), but not gender, were also associated with metformin initiation. Approximately 20% of metformin initiators had a second OH drug added within <18 months. While we were unable to ascertain specific contraindications to metformin (renal insufficiency, hepatic failure), <10% of the general population are expected to have these conditions.Seventy per cent of new initiators of OH treatment in Austria received metformin as recommended by international guidelines. At least 20% did not, taking into account possible contraindications, which provides an opportunity for intervention.
View details for DOI 10.1002/pds.2059
View details for Web of Science ID 000286071700008
View details for PubMedID 21182153
Primary Medication Non-Adherence: Analysis of 195,930 Electronic Prescriptions
JOURNAL OF GENERAL INTERNAL MEDICINE
2010; 25 (4): 284-290
Non-adherence to essential medications represents an important public health problem. Little is known about the frequency with which patients fail to fill prescriptions when new medications are started ("primary non-adherence") or predictors of failure to fill.Evaluate primary non-adherence in community-based practices and identify predictors of non-adherence.75,589 patients treated by 1,217 prescribers in the first year of a community-based e-prescribing initiative.We compiled all e-prescriptions written over a 12-month period and used filled claims to identify filled prescriptions. We calculated primary adherence and non-adherence rates for all e-prescriptions and for new medication starts and compared the rates across patient and medication characteristics. Using multivariable regressions analyses, we examined which characteristics were associated with non-adherence.Primary medication non-adherence.Of 195,930 e-prescriptions, 151,837 (78%) were filled. Of 82,245 e-prescriptions for new medications, 58,984 (72%) were filled. Primary adherence rates were higher for prescriptions written by primary care specialists, especially pediatricians (84%). Patients aged 18 and younger filled prescriptions at the highest rate (87%). In multivariate analyses, medication class was the strongest predictor of adherence, and non-adherence was common for newly prescribed medications treating chronic conditions such as hypertension (28.4%), hyperlipidemia (28.2%), and diabetes (31.4%).Many e-prescriptions were not filled. Previous studies of medication non-adherence failed to capture these prescriptions. Efforts to increase primary adherence could dramatically improve the effectiveness of medication therapy. Interventions that target specific medication classes may be most effective.
View details for DOI 10.1007/s11606-010-1253-9
View details for Web of Science ID 000275779300003
View details for PubMedID 20131023
A SAS macro for a clustered permutation test
COMPUTER METHODS AND PROGRAMS IN BIOMEDICINE
2009; 95 (1): 89-94
The clustered permutation test is a nonparametric method of significance testing for correlated data. It is often used in cluster randomized trials where groups of patients rather than individuals are randomized to either a treatment or control intervention. We describe a flexible and efficient SAS macro that implements the 2-sample clustered permutation test. We discuss the theory and applications behind this test as well as details of the SAS code.
View details for DOI 10.1016/j.cmpb.2009.02.005
View details for Web of Science ID 000266187900008
View details for PubMedID 19321221
The Relationship Between Focal Erosions and Generalized Osteoporosis in Postmenopausal Women With Rheumatoid Arthritis
ARTHRITIS AND RHEUMATISM
2009; 60 (6): 1624-1631
Among rheumatoid arthritis (RA) patients who have had the disease for 10 years, more than half have focal erosions, and the risk of fracture is doubled. However, there is little information about the potential relationship between focal erosions and bone mineral density (BMD). The aim of this study was to determine whether lower BMD is associated with higher erosion scores among patients with RA.We enrolled 163 postmenopausal women with RA, none of whom were taking osteoporosis medications. Patients underwent dual x-ray absorptiometry at the hip and spine and hand radiography, and completed a questionnaire. The hand radiographs were scored using the Sharp method, and the relationship between BMD and erosions was measured using Spearman's correlation coefficients and adjusted linear regression models.Patients had an average disease duration of 13.7 years, and almost all were taking a disease-modifying antirheumatic drug. Sixty-three percent were rheumatoid factor (RF) positive. The median modified Health Assessment Questionnaire score was 0.7, and the average Disease Activity Score in 28 joints was 3.8. The erosion score was significantly correlated with total hip BMD (r=-0.33, P<0.0001), but not with lumbar spine BMD (r=-0.09, P=0.27). Hip BMD was significantly lower in RF-positive patients versus RF-negative patients (P=0.02). In multivariable models that included age, body mass index, and cumulative oral glucocorticoid dose, neither total hip BMD nor lumbar spine BMD was significantly associated with focal erosions.Our results suggest that hip BMD is associated with focal erosions among postmenopausal women with RA, but that this association disappears after multivariable adjustment. While BMD and erosions may be correlated with bone manifestations of RA, their relationship is complex and influenced by other disease-related factors.
View details for DOI 10.1002/art.24551
View details for Web of Science ID 000267116800010
View details for PubMedID 19479876
An evaluation of statistical approaches for analyzing physician-randomized quality improvement interventions
CONTEMPORARY CLINICAL TRIALS
2008; 29 (5): 687-695
Health care quality improvement interventions are often evaluated in randomized trials in which individual physicians serve as the unit of randomization. These cluster randomized trials present a unique data structure that consists of many clusters of highly variable size. The appropriate method of analysis for these trials is unknown. We conducted a simulation study to compare several methods for analyzing data which were generated to replicate the structure of our motivating example. We varied the treatment effect size and the distributional assumptions about the random effect. Simulation was used to estimate power, coverage, bias, and mean squared error of full maximum likelihood estimation (MLE), approximate MLE using penalized quasi-likelihood (PQL), generalized estimating equations (GEE), group-bootstrapped logistic regression, and a clustered permutation test. Across all conditions tested, GEE and full MLE performed comparably. Bootstrapped methods were less powerful and had higher mean squared error under conditions of variable cluster size. PQL yielded biased results. The permutation test preserved Type I error rates, but had less power than the other methods considered.
View details for DOI 10.1016/j.cct.2008.04.003
View details for Web of Science ID 000259424400008
View details for PubMedID 18571476