Zhehui Luo

Applying Propensity Score Methods in Medical Research: Pitfalls and Prospects

The authors review experimental and nonexperimental causal inference methods, focusing on assumptions for the validity of instrumental variables and propensity score (PS) methods. They provide guidance in four areas for the analysis and reporting of PS methods in medical research and selectively evaluate mainstream medical journal articles from 2000 to 2005 in the four areas, namely, examination of balance, overlapping support description, use of estimated PS for evaluation of treatment effect, and sensitivity analyses. In spite of the many pitfalls, when appropriately evaluated and applied, PS methods can be powerful tools in assessing average treatment effects in observational studies. Appropriate PS applications can create experimental conditions using observational data when randomized controlled trials are not feasible and, thus, lead researchers to an efficient estimator of the average treatment effect.

Patterns of psychotropic medication use in very young children with attention-deficit hyperactivity disorder

ABSTRACT. Psychotropic medications are increasingly used for very young children. Patterns of use in a well-described group of children 3 years and younger with a diagnostic label of attention-deficit hyperactivity disorder (ADHD) reveal both reasons to use such medications and concerns about how these medications are used. Of 223 children with ADHD, more than half (n = 127) received psychotropic medications in an idiosyncratic manner, both in the specific medication and in use over time. Almost half of the children who were medicated did not have opportunities for monitoring as often as every 3 months, despite the fact that more than half received psychotropic medications for 6 months or longer. Children with comorbid mental health conditions and chronic health conditions were at greater risk for receiving psychotropic medications. These patterns of use demonstrate a compelling need for guidance in psychopharmacological treatment of very young children.

Effect of Pretreatment With Statins on Ischemic Stroke Outcomes

Background and Purpose— Statins reduce the risk of stroke in at-risk populations and may improve outcomes in patients taking statins before an ischemic stroke (IS). Our objectives were to examine the effects of pretreatment with statins on poor outcome in IS patients. Methods— Over a 6-month period all acute IS admissions were prospectively identified in 15 hospitals participating in a statewide acute stroke registry. Poor stroke outcome was defined as modified Rankin score ≥4 at discharge (ie, moderate-severe disability or death). Multivariable logistic regression models and matched propensity score analyses were used to quantify the effect of statin pretreatment on poor outcome. Results— Of 1360 IS patients, 23% were using statins before their stroke event and 42% had a poor stroke outcome. After multivariable adjustment, pretreatment with statins was associated with lower odds of poor outcome (OR=0.74, 95% CI 0.52, 1.02). A significant interaction (P<0.01) was found between statin use and race. In whites, statins were associated with statistically significantly lower odds of poor outcome (OR=0.61, 95% CI 0.42, 0.86), but in blacks statins were associated with a nonstatistically significant increase in poor outcome (OR=1.82, 95% CI 0.98, 3.39). Matched propensity score analyses were consistent with the multivariable model results. Conclusions— Pretreatment with statins was associated with better stroke outcomes in whites, but we found no evidence of a beneficial effect of statins in blacks. These findings indicate the need for further studies, including randomized trials, to examine differential effects of statins on ischemic stroke outcomes among whites and blacks.

Alleviating perinatal depressive symptoms and stress: a nurse-community health worker randomized trial

To determine whether a Nurse-Community Health Worker (CHW) home visiting team, in the context of a Medicaid enhanced prenatal/postnatal services (EPS), would demonstrate greater reduction of depressive symptoms and stress and improvement of psychosocial resources (mastery, self-esteem, social support) when compared with usual Community Care (CC) that includes Medicaid EPS delivered by professionals. Greatest program benefits were expected for women who reported low psychosocial resources, high stress, or both at the time of enrollment. Medicaid eligible pregnant women (N = 613) were randomly assigned to either usual CC or the Nurse-CHW team. Mixed effects regression was used to analyze up to five prenatal and postnatal psychosocial assessments. Compared to usual CC, assignment to the Nurse-CHW team resulted in significantly fewer depressive symptoms, and as hypothesized, reductions in depressive symptoms were most pronounced for women with low psychosocial resources, high stress, or both high stress and low resources. Outcomes for mastery and stress approached statistical significance, with the women in the Nurse-CHW group reporting less stress and greater mastery. Women in the Nurse-CHW group with low psychosocial resources reported significantly less perceived stress than women in usual CC. No differences between the groups were found for self-esteem and social support. A Nurse-CHW team approach to EPS demonstrated advantage for alleviating depressive symptoms in Medicaid eligible women compared to CC, especially for women at higher risk.

Medicaid, Medicare, and the Michigan Tumor Registry: A Linkage Strategy

The study of health outcomes and the reduction in health disparities is at the forefront of the nations health care agenda. A theme in the disparities literature is the call for a data infrastructure that can track progress toward goals aimed at reducing differences in health outcomes. This article describes a strategy for linking Medicaid, Medicare, and Michigan Tumor Registry data for the purposes of studying disparities in cancer diagnosis, quality of care, and survival. The authors review their procedures for ensuring that a correct match between files occurred and offer guidance for merging and assessing the quality of these complex linked data sets. A cohort of 113,604 subjects (90%) from a population of 125,900 subjects was correctly linked from the Michigan Tumor Registry to Medicare and Medicaid files. Using probabilistic and deterministic methods, the prediction rate of the Medicaid match to the Michigan Tumor Registry was 93%. Approximately 13% of the subjects were dually eligible for Medicare and Medicaid. An expansive data set reflecting the Medicare and Medicaid medical service utilization and outcomes for a cohort of individuals age 65 years and older when diagnosed with cancer was created. This data set serves as a cornerstone of a health outcomes data infrastructure. The methodology described may serve as a model for other researchers seeking to create a similar data set in their state.

Influence of predispositions on post-traumatic stress disorder: does it vary by trauma severity?

BACKGROUND Only a minority of trauma victims (<10%) develops post-traumatic stress disorder (PTSD), suggesting that victims vary in predispositions to the PTSD response to traumas. It is assumed that the influence of predispositions is inversely related to trauma severity: when trauma is extreme predispositions are assumed to play a secondary role. This assumption has not been tested. We estimate the influence of key predispositions on PTSD induced by an extreme trauma - associated with a high percentage of PTSD - (sexual assault), relative to events of lower magnitude (accidents, disaster, and unexpected death of someone close). METHOD The National Epidemiologic Survey on Alcohol and Related Conditions (NESARC) is representative of the adult population of the USA. A total of 34 653 respondents completed the second wave in which lifetime PTSD was assessed. We conducted three series of multinomial logistic regressions, comparing the influence of six predispositions on the PTSD effect of sexual assault with each comparison event. Three pre-existing disorders and three parental history variables were examined. RESULTS Predispositions predicted elevated PTSD risk among victims of sexual assault as they did among victims of comparison events. We detected no evidence that the influence of predispositions on PTSD risk was significantly lower when the event was sexual assault, relative to accidents, disasters and unexpected death of someone close. CONCLUSIONS Important predispositions increase the risk of PTSD following sexual assault as much as they do following accidents, disaster, and unexpected death of someone close. Research on other predispositions and alternative classifications of event severity would be illuminating.

Primary care physicians treat somatization

BACKGROUNDWe hypothesized that somatizing patients managed by primary care physicians (PCP) would improve with a relationship-based intervention.METHODSWe randomized 30 adults with medically unexplained symptoms to treatment or usual care. Four PCPs were trained to intervene with cognitive-behavioral, pharmacological, and patient-centered management and deployed the intervention with seven scheduled visits over 12 months. Outcomes obtained at baseline and 12 months were: Mental component summary (MCS), the primary endpoint, and measures of physical and psychological symptoms and of satisfaction with the PCP.RESULTSPatients averaged 52.5 years; 83.3% were female; 79.6% were black. Using a difference of differences approach, we found that the intervention produced a large effect size (ES) (0.82; CI: 0.08 to 1.57) for the MCS in the predicted direction, similar to the ES for physical (−0.80; CI: −1.55 to −0.04) and psychological (−1.06; CI: −1.83 to −0.28) improvement and for increased satisfaction with the PCP (0.94; CI: 0.15 to 1.74). Using ANCOVA in a sensitivity analysis, we found that the ES fell slightly (0.59), while other measures were unchanged.CONCLUSIONSModerate-large effect sizes support the hypothesis that PCPs can effectively treat somatization. This points to the importance of performing a full RCT.

Diagnosis of advanced cancer among elderly medicare and medicaid patients

Background:Medicaid is implicated in late-stage cancer diagnoses, which is the primary indicator of a poor prognosis. Objective:We examined Medicaid enrollment and cancer diagnosis in patients ages 66 years and older. Medicaid enrollment was defined as enrolled 12+ months before diagnosis, enrolled <12 months before diagnosis, and enrolled after diagnosis. Subjects:Medicaid and Medicare administrative data were merged with the Michigan Tumor Registry to extract a sample of 46,109 patients with a first primary diagnosis of prostate, lung, breast, or colorectal cancer between 1997 and 2000. Measures were: (1) diagnosed during the same month as death; (2) invasive, but unknown stage; and (3) regional or distant stage disease. Results:Patients enrolled in Medicaid <12 months before diagnosis were at greater risk of breast (odds ratio [OR] = 2.70; 95% confidence interval [95% CI] = 1.22–5.99) and lung (OR = 2.18; 95% CI = 1.45–3.29) cancer diagnosis in the month of death than Medicare only patients. Similarly, patients with a history of Medicaid enrollment had a high risk of diagnosis with invasive, but unknown breast, lung, and prostate cancer stage. Patients enrolled in Medicaid following diagnosis had a higher risk of late stage colorectal (OR = 1.30; 95% CI = 1.01–1.67), breast (OR = 2.12; 95% CI = 1.60–2.82), and lung (OR = 1.33; 95% CI = 1.02–1.75) cancer relative to Medicare only patients. Conclusions:There is a preponderance of cancer diagnosis at death and cancer diagnosis with invasive but unknown stage in the Medicaid population, but the appropriateness of these diagnoses is unclear. Late-stage cancer tends to precipitate Medicaid enrollment.

Combining Registry, Primary, and Secondary Data Sources to Identify the Impact of Cancer on Labor Market Outcomes

Some cancers are rapidly becoming chronic conditions that are more and more often diagnosed in working-age individuals. The authors developed a research agenda to study the labor market outcomes attributable to detection and treatment for cancer, and research design and data collection strategies to improve upon other research on these questions. In this article, they describe their approach to combining secondary data sources, primary data collection, and cancer registry data to evaluate the impact cancer has on labor market outcomes such as employment, hours worked, wages, and health insurance. They then critically assess how well their study design and data collection strategy accomplished its objectives. The intention is to offer guidance on how researchers, who are interested in the economic consequences of cancer, as well as of other chronic conditions, might develop and execute studies that examine labor market outcomes. As more attention is placed on the economic aspects of disease, the methods used to estimate productivity loss and other economic outcomes attributable to these conditions require careful scrutiny so that reliable findings can be used to shape health care decisions and policy.

Occurrence of lead-related symptoms below the current Occupational Safety and Health Act allowable blood lead levels

To determine the occurrence of symptoms of lead toxicity at levels below the current allowable Occupational Safety and Health Act blood lead level of 50 &mgr;g/dL, standardized telephone interviews were conducted of individuals reported to a statewide laboratory-based surveillance system. Four hundred and ninety-seven, or 75%, of the eligible participants were interviewed. Gastrointestinal, musculoskeletal, and nervous system symptoms increased with increasing blood lead levels. Nervous, gastrointestinal, and musculoskeletal symptoms all began to be increased in individuals with blood leads between 30–39 &mgr;g/dL and possibly at levels as low as 25–30 &mgr;g/dL for nervous system symptoms. The results of this study of increased symptoms are consistent with and provide added weight to previous results showing subclinical changes in the neurologic and renal systems and sperm counts at blood lead levels currently allowed by the Occupational Safety and Health Act.