Clinical and Informatics Research on Large Clinical Databases

National Library Of Medicine

Investigators

Linked publications & trials

Abstract

A. Possible Beneficial and Adverse Drug Effects on SARS-COV-2 Infections (COVID-19) We have identified one set of drugs used in ambulatory care that might reduce or increase incidence and/or severity of COVID-19. This combination is being tested in clinical trials, but the number of subjects has been small and observational data will likely be needed to support these studies. Some drugs that might reduce or aggravate COVID-19 include non-steroidals, anticoagulants, H2 blockers, anti-platelet agents, ACE inhibitors, and ARBs that treat rheumatologic disease (Tocilizumab, Rituximab, Sarilumab). We will study the association with the incidence and severity of COVID-19 infections. B. Long COVID syndrome in Elderly Patients Incidence of long COVID in the elderly is difficult to estimate because the long COVID may be masked by existing chronic diseases and thus they can be under-reported. We estimate the incidence of long COVID among elderly Medicare beneficiaries over 65 by a constellation of 11 symptoms, the WHOs clinical definition. With the same set of symptoms, we identify cases of postulated long Flu. Then we compare long COVID with long Flu in terms of incidence, symptomatology, and healthcare utilization. C. Maintenance drugs and longevity A large proportion of elderly Medicare beneficiaries aged 65 and over suffer from hypertension and hyperlipidemia. They are treated with multiple maintenance drugs according to whether the condition is chronic. Studies that reported the efficacy of a maintenance drug did not include multiple drugs that were taken simultaneously by real-life patients in one unified analysis. An effect of one drug could be confounded with that of other drugs and thus conclusions often contradict. We include 6 maintenance drugs: 5 antihypertensive medications (diuretics, beta-blockers, calcium-channel blockers, ACE inhibitors, and ARBs) and 1 lipid-lowering drug (statin). We further separate some antihypertensive drug classes into subclasses to differentiate varieties of indication within a class and antihypertensive potency: 2 subclasses of calcium-channel blockers, 7 subclasses of diuretics and 3 subclasses of beta-blockers. Then, we report comparative effectiveness of each of 15 maintenance drugs on all-cause mortality among 2 million elderly Medicare beneficiaries. D. Proton Pump Inhibitors (PPIs) and mortality PPIs have been associated with increases in the incidence of pneumonia, C Decile infection and osteoporosis/fractures, probable chronic renal failure, and cardiac events. Xie et al reported that PPIs could also increase the risk of death using Veterans Affairs (VA) data (HR of 1.15 1.25). From 2007-2016 Medicare Parts A, B and D data, we identified 1.2 million Medicare beneficiaries. We also defined a set of covariates: use of PPIs, use of H2 blockers as a control drug, admission to intensive care units or inpatient hospitals, socio-demographics, presence of 58 chronic conditions and treated them as time-dependent covariates in our main analysis of Cox proportion hazard regression. In addition to treating covariates in time-varying manner, we used the concept of lag-time to define drug exposure period in order to control for protopathic bias which occurs when the outcome of interest is associated with an exposure that actually results from early signs and symptoms of the outcome under study. With use of lag times, PPIs had no associations with death, in agreement with one RCT that showed no such association. The Editor-in-Chief of Clinical Gastroenterology and Hepatology has identified our study as being highly relevant to individuals who are involved with patients care and invited us in a new publication called the AGA Clinicians Companion to highlight it. E. Association between Female Hormone replacement therapy (HRT) and longevity, cardiovascular diseases and cancers HRT is an effective treatment for the typical menopause-related symptoms (such as hot flashes, night sweats, irregular periods etc.) and long-term health problems associated with menopause (the risk of osteoporosis, cardiovascular disease and stroke). Reports of some studies have trumpeted negative effects of HRT on outcomes such as cardiovascular diseases, cancers, and all-cause mortality. However, the evidence behind them is weak or has been reversed. In this study, we traced about 1.5 million female Medicare beneficiaries from Medicare Part D entry to the onset of each outcome, death, switching to capitated plan, disenrollment from Medicare, or end of data availability whichever comes first and then we compared each risk among women treated with HRT of various kinds with to those not treated, and we treated almost all covariates as time time-dependent in a Cox proportional hazard regression analysis. Estrogen use, but not combined estrogen+progestin use, was associated with less risk of breast and other studied cancers and a significant reduction in mortality risk. F. Evaluating the risk of fractures among elderly women enrolled in Medicare Osteoporosis characterized by progressive deterioration of bone structure due to decreased bone mineral density (BMD) has been found to be closely associated with fractures. There are several pharmacotherapies available for prevention and treatment of postmenopausal osteoporosis including bisphosphonate, estrogen, raloxifene, denosumab and more. However, their beneficial and/or detrimental effect on fractures is not well addressed. We are conducting a nationwide cohort study of patients with osteoporosis to compare risks of any, hip and atypical femur fractures among patients treated with any of the drugs. G. Developing a method to combine machine learning methodology with time-to-event data analysis. Several new methods for survival analysis have been proposed by extending the Cox proportional hazard model with neural networks. They were compared to classical Cox regression analysis and were found to be highly competitive, yielding the best performance in terms of Brier score and binomial log-likelihood. Though classical Cox regression analysis yields instantaneous hazards based on covariates that change over time during the follow-up period, none of newly proposed methods do not incorporate the effects of time-varying covariates. We are updating a currently available Python package combining Cox model with neural networks, a.k.a PyCOX, to incorporate time-varying covariates and comparing it with a conventional Cox regression with time-varying covariates in terms of hazard ratios and the concordance index using a real-word Medicare data.