Search PubMed⌕ Search

Biomedical subjects

P R Rosenbaum

Publications and source records attributed to P R Rosenbaum.

At least 19 recordsLinked to original sources
PubMed2001

Multivariate matching and bias reduction in the surgical outcomes study.

BACKGROUND: Outcomes studies often need a level of detail that is not present in administrative data, therefore requiring abstraction of medical charts. Case-control methods may be used to improve statistical power and reduce abstraction costs, but limitations of exact matching often preclude the use of many covariates. Unlike exact matching, multivariate matching may allow cases to be matched simultaneously on hundreds of covariates. OBJECTIVES: To develop multivariate matched case-control pairs in a study of death after surgery in the Medicare population. RESEARCH DESIGN: Using 830 randomly selected index cases of patients who died within 60 days from admission, controls were found who did not die within that time period, matching on risk for death and other patient characteristics with up to 173 variables used simultaneously in the matching algorithms. SUBJECTS: General and orthopedic Medicare surgical cases in Pennsylvania from 1995 to 1996. Controls were either selected from across the entire state (108,765 possible subjects), or from within the same hospital as the case. MEASURES: Percent bias reduction and the average difference between cases and controls in units of standard deviations. RESULTS: Matched controls were far more similar to cases (deaths) upon admission to the hospital than typical patients, both in statewide and within hospital matches. Bias reduction was usually greater than 50% and often approached 100%. The difference between cases and matched controls for most variables was usually below 0.2 SD. CONCLUSIONS: Multivariate matching methods may aid in conducting studies with Medicare claims records by improving the quality of matches, thereby achieving a better understanding of the etiology of outcomes.

Aged↗
PubMed2000

Substantial gains in bias reduction from matching with a variable number of controls.

In observational studies that match several controls to each treated subject, substantially greater bias reduction is possible if the number of controls is not fixed but rather is allowed to vary from one matched set to another. In certain cases, matching with a fixed number of controls may remove only 50% of the bias in a covariate, whereas matching with a variable number of controls may remove 90% of the bias, even though both control groups have the same number of controls in total. An example of matching in a study of surgical mortality is discussed in detail.

Bias↗
PubMed1999

Invited commentary: propensity scores.

The propensity score is the conditional probability of exposure to a treatment given observed covariates. In a cohort study, matching or stratifying treated and control subjects on a single variable, the propensity score, tends to balance all of the observed covariates; however, unlike random assignment of treatments, the propensity score may not also balance unobserved covariates. The authors review the uses and limitations of propensity scores and provide a brief outline of associated statistical theory. They also present a new result of using propensity scores in case-cohort studies.

Case-Control Studies↗
PubMed1999

Reduced sensitivity to hidden bias at upper quantiles in observational studies with dilated treatment effects.

When a treatment has a dilated effect, with larger effects when responses are higher, there can be much less sensitivity to bias at upper quantiles than at lower quantiles; i.e., small, plausible hidden biases might explain the ostensible effect of the treatment for many subjects, and yet only quite large hidden biases could explain the effect on a few subjects having dramatically elevated responses. An example concerning kidney function of cadmium workers is discussed in detail. In that example, the treatment effect is far from additive: It is plausibly zero at the lower quartile of responses to control, and it is large and fairly insensitive to bias at the upper quartile.

Bias↗
PubMed1999

Conditional Length of Stay.

OBJECTIVE: To develop and test a new outcome measure, Conditional Length of Stay (CLOS), to assess hospital performance when deaths are rare and complication data are not available. DATA SOURCES: The 1991 and 1992 MedisGroups National Comparative Data Base. STUDY DESIGN: We use engineering reliability theory traditionally applied to estimate mechanical failure rates to construct a CLOS measure. Specifically, we use the Hollander-Proschan statistic to test if LOS distributions display an "extended" pattern of decreasing hazards after a transition point, suggesting that "the longer a patient has stayed in the hospital, the longer a patient will likely stay in the hospital" versus an alternative possibility that "the longer a patient has stayed in the hospital, the faster a patient will likely be discharged from the hospital." DATA COLLECTION/EXTRACTION METHODS: Abstracted records from 7,777 pediatric pneumonia cases and 3,413 pediatric appendectomy cases were available for analysis. PRINCIPAL FINDINGS: For both conditions, the Hollander-Proschan statistic strongly displays an "extended" pattern of LOS by day 3 (p<.0001) associated with declining rates of discharge. This extended pattern coincides with increasing patient complication rates. Worse admission severity and chronic disease contribute to lower rates of discharge after day 3. CONCLUSIONS: Patient stays tend to become prolonged after complications. By studying CLOS, one can determine when the rate of hospital discharge begins to diminish--without the need to directly observe complications. Policymakers looking for an objective outcome measure may find that CLOS aids in the analysis of a hospital's management of complicated patients without requiring complication data, thereby facilitating analyses concerning the management of patients whose care has become complicated.

Appendectomy↗
PubMed1997

The relationship between choice of outcome measure and hospital rank in general surgical procedures: implications for quality assessment.

OBJECTIVE: Institutional complication rates are often used to assess hospital quality of care, particularly for conditions and procedures where mortality rates are not useful because deaths are rare. The objective of this study was to assess the correlation among hospital quality assessment rankings based on adjusted mortality, complication and failure-to-rescue rates. DESIGN: This study used a clinically detailed administrative data set to compare severity and case-mix adjusted hospital outcome rankings for three different measures of quality of care: in-hospital death, complication and failure-to-rescue (in-hospital death following a complication). SETTING AND PATIENTS: Analysis of 74,647 patients who underwent general surgical procedures included in the 1991 and 1992 MedisGroups National Comparative Data Base. MEASUREMENTS: Adjusted outcomes of death, complication and failure to rescue based on multivariable logistic regression models. RESULTS: For 142 hospitals, the correlation between hospital rankings based on the death rate and those ranked by the complication rate was only 0.208 (P = 0.013). A similarly low correlation was present between the complication and failure rate rankings, r = -0.090 (P = 0.287). A higher correlation was observed between the death and failure rate rankings, r = 0.90 (P < 0.001). CONCLUSIONS: For general surgical procedures, hospital rank using the complication rate is poorly correlated with rankings using the death or failure rate. Complication rates should be used with great caution and should not be used in isolation when assessing hospital quality of care.

Hospital Mortality↗
PubMed1997

A spurious correlation between hospital mortality and complication rates: the importance of severity adjustment.

OBJECTIVES: When two outcome measures, such as mortality and complication rates, are intended to measure the same underlying quantity (in this case hospital quality of care), one expects they will be highly correlated. In addition, as data quality improves, one expects the correlation will increase. The authors show that these expectations are, in a significant way, mistaken. METHODS: The authors study two outcomes (hospital mortality and complication rates after surgery) using three predictive models that vary in adjustment for severity of illness. RESULTS: Two hospital rankings, based on each of the two outcomes, are well correlated when not adjusted for severity. However, as clinical data are added to the models, the correlation tends to disappear. The authors explain this based on assumptions regarding the relative size of the partial correlations between mortality, complication rate, and severity covariates. CONCLUSIONS: Before claims of construct validity can be made, investigators must show that correlations between outcomes purporting to measure quality of care are sustained after adequate correction for severity. Most importantly, it should be recognized that inadequately controlled confounding variables may lead to a spurious high correlation between an accepted and a new outcome measure, and a false sense of adequate construct validity.

Adult↗
PubMed1997

Signed rank statistics for coherent predictions.

A generalization of Wilcoxon's signed rank test is proposed for testing a dose-response relationship with one or more outcomes. The test is useful in matched observational studies or in nonrandomized experiments that use dose-response relationships and predictions about multiple outcomes in an effort to distinguish actual treatment effects from hidden biases. A sensitivity analysis indicating whether a dose-response relationship or multiple predictions are confirmed with sufficient strength to reduce sensitivity to hidden bias is performed. Together, the test and the sensitivity analysis help to quantify the degree to which a coherent pattern of associations is present or absent, and the degree to which this strengthens or fails to strengthen evidence of cause and effect. Formal properties of tests of this kind are examined. The form of the optimal test is determined, though this test is not usable because it depends upon the values of the unknown parameters under study. Also examined are the conditions under which the proposed test resembles the optimal test, as well as the impact of various violations of those conditions on power. An example involving matched pairs exposed to varying doses of cadmium is considered in detail.

Bias↗
PubMed1995

Evaluation of the complication rate as a measure of quality of care in coronary artery bypass graft surgery.

OBJECTIVE: To determine whether hospital rankings based on complication rates provide the same information as hospital rankings based on mortality rates. DESIGN: A retrospective study of in-hospital death, complication, and death following complication (failure to rescue). Hospitals were ranked using residuals based on the difference between the observed and the expected number of events (from logistic regression models); rankings were compared using Spearman rank correlations. SETTING: Hospitals performing coronary artery bypass graft (CABG) surgery in the 1991 and 1992 MedisGroups National Comparative Data Bases. PATIENTS AND DATA SETS: Record abstraction data for 16,673 patients who underwent CABG procedures at 57 hospitals, linked with data from the 1991 American Hospital Association Annual Survey. RESULTS: After adjusting for patient admission severity of illness, there were low correlations between hospital rankings based on death or failure to rescue and those rankings based on complication (death vs complication, r = 0.07, P = .58; failure to rescue vs complication, r = -0.22, P = .11). In addition, many hospital characteristics that are generally associated with a higher quality of care were associated with higher complication rates but with expected or lower-than-expected mortality rates. CONCLUSIONS: Hospital rankings based on complication rates provide different information than those based on mortality rates. Until more is known about these differences, complication rates should not be used to judge hospital quality of care in CABG surgery.

Coronary Artery Bypass↗
PubMed1994

Coherence in observational studies.

It is often said that the coherence of an association between a treatment and outcomes is important in judging whether the association is causal. An attempt is made to quantify the evidence provided by a coherent association. This is done in two steps. First, a test is developed to detect a coherent association. The test is a generalization of the tests of Mann and Whitney, Wilcoxon, and Gehan. Second, the sensitivity of the test to hidden bias is examined. The question is whether a coherent pattern of associations implies less sensitivity to hidden biases. An example is considered in detail.

Animals↗
PubMed1991

Discussing hidden bias in observational studies.

In observational studies or nonrandomized experiments, treated and control groups may differ in their outcomes even if the treatment has no effect; this may happen if the groups were not comparable before the start of treatment. The groups may fail to be comparable in either of two ways: They may differ with respect to characteristics that have been measured, in which case there is an overt bias, or they may differ in ways that have not been measured, in which case there is a hidden bias. Overt biases are controlled through adjustments, such as matching. Hidden bias is more difficult to address because the relevant measurements are not available. A sensitivity analysis asks how much hidden bias would need to be present if hidden bias were to explain the differing outcomes in the treated and control groups. A sensitivity analysis provides a tangible and specific framework for discussing hidden biases.

Bias↗
PubMed1991

Sensitivity analysis for matched case-control studies.

A sensitivity analysis in an observational study indicates the degree to which conclusions would be altered by hidden biases of various magnitudes. A method of sensitivity analysis previously proposed for cohort studies is extended for use in matched case-control studies with multiple controls, where slightly different derivations and calculations are required. Also discussed is a sensitivity analysis for case-control studies that have two distinct types of controls, say hospital and neighborhood controls, where the two types may be affected by different biases. For illustration, the method is applied to five case-control studies, including a study of herniated lumbar disc in which there are three types of cases, and a study of breast cancer with two types of controls.

Biometry↗
PubMed1985

The bias due to incomplete matching.

Observational studies comparing groups of treated and control units are often used to estimate the effects caused by treatments. Matching is a method for sampling a large reservoir of potential controls to produce a control group of modest size that is ostensibly similar to the treated group. In practice, there is a trade-off between the desires to find matches for all treated units and to obtain matched treated-control pairs that are extremely similar to each other. We derive expressions for the bias in the average matched pair difference due to the failure to match all treated units--incomplete matching, and the failure to obtain exact matches--inexact matching. A practical example shows that the bias due to incomplete matching can be severe, and moreover, can be avoided entirely by using an appropriate multivariate nearest available matching algorithm, which, in the example, leaves only a small residual bias due to inexact matching.

Analysis of Variance↗
PubMed1984

Aggressive combined modality therapy for advanced local-regional breast carcinoma.

Thirty-two women with advanced local regional breast carcinoma, including nine patients with histologically diagnosed inflammatory cancer, were entered on a prospective pilot study. They were treated aggressively with initial surgery, two courses of induction chemotherapy with cyclophosphamide, methotrexate, 5-fluorouracil, +/- prednisone, +/- tamoxifen (CMF [P] [T]), local-regional radiotherapy, and then maintenance chemotherapy with CMF(P) (T) alternating with doxorubicin, vincristine, +/- tamoxifen (AV[T]). The patients have been followed for 19-70 months from the time of mastectomy and their actuarial three-year survival is 65% with a median survival that has not yet been reached. Median disease-free survival (time to progression) is currently 29.5 months. Women whose gross disease could not be totally resected surgically had shorter disease-free survivals than those rendered surgically free of disease (p = 0.01). Clinically evident cardiotoxicity was seen in 25% of the patients and was felt to be primarily due to the combination of doxorubicin and radiation therapy. It was significantly more common (Plt less than 0.05) in patients with left chest irradiation (seven of 18 women) as opposed to those with right-sided irradiation (one of 14).

Adult↗
PubMed1984

Difficulties with regression analyses of age-adjusted rates.

A common type of observational study compares population rates in several regions having differing policies in an effort to assess the effects of those policies. In many studies, particularly in public health and epidemiology, age-adjusted rates are regressed on predictor variables to give a covariance-adjusted estimate of effect; this estimate is shown to be generally biased for the appropriate regression coefficient. For familiar models, the analysis of crude rates with age as a covariate can lead to unbiased estimates, and therefore can be preferable. Several other regression methods are also considered.

Age Factors↗