Matching One Sample According to Two Criteria in Observational Studies.

Multivariate matching has two goals: (i) to construct treated and control groups that have similar distributions of observed covariates, and (ii) to produce matched pairs or sets that are homogeneous in a few key covariates. When there are only a few binary covariates, both goals may be achieved by matching exactly for these few covariates. Commonly, however, there are many covariates, so goals (i) and (ii) come apart, and must be achieved by different means. As is also true in a randomized experiment, similar distributions can be achieved for a high-dimensional covariate, but close pairs can be achieved for only a few covariates. We introduce a new polynomial-time method for achieving both goals that substantially generalizes several existing methods; in particular, it can minimize the earthmover distance between two marginal distributions. The method involves minimum cost flow optimization in a network built around a tripartite graph, unlike the usual network built around a bipartite graph. In the tripartite graph, treated subjects appear twice, on the far left and the far right, with controls sandwiched between them, and efforts to balance covariates are represented on the right, while efforts to find close individual pairs are represented on the left. In this way, the two efforts may be pursued simultaneously without conflict. The method is applied to our on-going study in the Medicare population of the relationship between superior nursing and sepsis mortality. The match2C package in R implements the method.

Mortality and cardiovascular disease among older live kidney donors.

Over the past two decades, live kidney donation by older individuals (≥55 years) has become more common. Given the strong associations of older age with cardiovascular disease (CVD), nephrectomy could make older donors vulnerable to death and cardiovascular events. We performed a cohort study among older live kidney donors who were matched to healthy older individuals in the Health and Retirement Study. The primary outcome was mortality ascertained through national death registries. Secondary outcomes ascertained among pairs with Medicare coverage included death or CVD ascertained through Medicare claims data. During the period from 1996 to 2006, there were 5717 older donors in the United States. We matched 3368 donors 1:1 to older healthy nondonors. Among donors and matched pairs, the mean age was 59 years; 41% were male and 7% were black race. In median follow-up of 7.8 years, mortality was not different between donors and matched pairs (p = 0.21). Among donors with Medicare, the combined outcome of death/CVD (p = 0.70) was also not different between donors and nondonors. In summary, carefully selected older kidney donors do not face a higher risk of death or CVD. These findings should be provided to older individuals considering live kidney donation.

Pediatric care for preterm infants after NICU discharge: high number of office visits and prescription medications.

OBJECTIVE: To determine the frequency and risk factors for clinic and pharmacy use in preterm infants during the first year after neonatal intensive care unit (NICU) discharge. STUDY DESIGN: We analyzed clinic visits and prescriptions in a cohort of 23 to 32 weeks infants. We constructed multivariable regression models to determine risk factors for high use. RESULT: The 892 preterm infants experienced 18 346 pediatric visits (mean 20 visits per infant per year) and filled 2100 prescriptions (mean 5.5 prescriptions per year among infants taking medications). Most visits were non-well child care visits to pediatric primary care providers. Prematurity and bronchopulmonary dysplasia (BPD) are important risk factors: infants at 23 to 26 weeks gestation or infants with BPD had an average 29 visits per year and 9 prescriptions per year among infants taking medication. However, half of the highest using infants were relatively healthy infants at 27 to 32 weeks gestation who escaped BPD, NEC or grade 3 to 4 intraventricular hemorrhage. CONCLUSION: Premature infants had frequent pediatric visits and prescription medications. Extreme prematurity and neonatal morbidities are important risk factors; however, half of the highest using infants are moderately preterm without neonatal morbidities.

Predicting cisplatin ototoxicity in children: the influence of age and the cumulative dose.

The aim of this study was to determine the risk factors for high-frequency hearing loss in children treated with cisplatin. We scored off-treatment pure-tone audiograms from 153 children (age 6 months to 18 years) who had completed cisplatin therapy (40-200 mg/m(2)/cycle) for germ cell tumours, hepatoblastoma, neuroblastoma or osteosarcoma. The risk of developing bilateral moderate to severe high-frequency hearing loss was significantly related to the age at treatment (P<0.001), and individual and cumulative cisplatin dosages (both P<0.005). Logistic regression showed that children younger than 5 years were at a greater risk of sustaining cisplatin ototoxicity than children older than 15 years, controlling for individual and cumulative doses of cisplatin (Odds Ratio (OR)=21.17, 95% Confidence Interval (CI): 2.48-180.94). Age at treatment and the cumulative dose of cisplatin were the two most important risk factors in predicting moderate to severe high-frequency hearing loss in children treated with cisplatin.

Design and baseline characteristics for the ACE Inhibitor After Anthracycline (AAA) study of cardiac dysfunction in long-term pediatric cancer survivors.

PURPOSE: The ACE Inhibitor After Anthracycline (AAA) study is a randomized, double-blind, controlled clinical trial comparing enalapril with placebo to determine whether treatment can slow the progression of cardiac decline in patients who screen positive for anthracycline cardiotoxicity. METHODS: The primary outcome measure is the rate of decline, over time, in maximal cardiac index (in liters per minute per meters squared) at peak exercise; the secondary outcome measure is the rate of increase in left ventricular end systolic wall stress (in grams per centimeters squared). Patients >2 years off therapy and <4 years from diagnosis, aged 8 years and older, were eligible if they had received anthracyclines and had at least one cardiac abnormality identified at any time after anthracycline exposure. RESULTS: A total of 135 patients were randomized to enalapril or placebo. Baseline characteristics were similar across treatment groups. CONCLUSIONS: The AAA study will provide important information concerning the efficacy of using angiotensin-converting enzyme inhibitors to offset the effects of late anthracycline cardiotoxicity.

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.

Matching and thick description in an observational study of mortality after surgery.

Multivariate matching permits the construction of matched pairs or matched sets that balance large numbers of observed covariates. Unlike model-based adjustments, in matching a patient remains intact as a single patient, and may be scrutinized as an individual and thickly described. A thick description entails a detailed, perhaps narrative, account of a patient's care, for instance, the account one might find in the 'Case Reports from the Massachusetts General Hospital' as published in the New England Journal of Medicine. While discussing certain principles of thick description, we illustrate using data from the pilot for a case-control study of the causes of death following surgery. Matching is based on billing data from Medicare, and the medical charts of matched pairs are then abstracted. In the pilot, we matched cases and controls in one hospital, located and scrutinized their medical charts. As a consequence, we corrected our misinterpretations of aspects of Medicare billing data, thereby improving the matching for the full study. Also, looking at charts suggested topics for investigation and helped us understand the types of information we might reliably find in charts, and this reshaped our plans for chart abstraction. Our central claim is that, unlike other methods of adjustment, matching facilitates thick description of a handful of cases, and such scrutiny of cases benefits statistical studies at several stages. Thick description of a few matched cases is used repeatedly to improve the matching and to design further data collection for the matched sample. Thick description aids matching by providing close examination of what the matching has actually accomplished, an examination that uses much more information than is available for use in matching. Matching aids thick description by placing side by side two patients who are fairly comparable, so that a thick description of them may usefully be performed.

Anesthesiologist direction and patient outcomes.

BACKGROUND: Anesthesia services for surgical procedures may or may not be personally performed or medically directed by anesthesiologists. This study compares the outcomes of surgical patients whose anesthesia care was personally performed or medically directed by an anesthesiologist with the outcomes of patients whose anesthesia care was not personally performed or medically directed by an anesthesiologist. METHODS: Cases were defined as being either "directed" or "undirected," depending on the type of involvement of the anesthesiologist, as determined by Health Care Financing Administration billing records. Outcome rates were adjusted to account for severity of disease and other provider characteristics using logistic regression models that included 64 patient and 42 procedure covariates, plus an additional 11 hospital characteristics often associated with quality of care. Medicare claims records were analyzed for all elderly patients in Pennsylvania who underwent general surgical or orthopedic procedures between 1991-1994. The study involved 194,430 directed and 23,010 undirected patients among 245 hospitals. Outcomes studied included death rate within 30 days of admission, in-hospital complication rate, and the failure-to-rescue rate (defined as the rate of death after complications). RESULTS: Adjusted odds ratios for death and failure-to-rescue were greater when care was not directed by anesthesiologists (odds ratio for death = 1.08, P < 0.04; odds ratio for failure-to-rescue = 1.10, P < 0.01), whereas complications were not increased (odds ratio for complication = 1.00, P < 0.79). This corresponds to 2.5 excess deaths/1,000 patients and 6.9 excess failures-to-rescue (deaths) per 1,000 patients with complications. CONCLUSIONS: Both 30-day mortality rate and mortality rate after complications (failure-to-rescue) were lower when anesthesiologists directed anesthesia care. These results suggest that surgical outcomes in Medicare patients are associated with anesthesiologist direction, and may provide insight regarding potential approaches for improving surgical outcomes. (Key words: Anesthesiologists; anesthesia care team; quality of care; mortality; failure-to-rescue; complication; Medicare; general surgery; orthopedics.)

Anesthesiologist direction and patient outcomes.

The relationship between physicians and nurses in the delivery of anesthesia care is politically and financially charged, and hotly debated. Against this backdrop, federal regulators have proposed dropping a Medicare requirement that nurse anesthetists be supervised by a physician. Proponents note that the new regulations would resolve inconsistencies between Medicare supervisory requirements and state law, while opponents voice concerns for patient safety. This Issue Brief describes the current controversy, and summarizes a newly published study that suggests differences in patient outcomes depending on the nature and level of anesthesiologist involvement in surgical care.

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.

The influence of chronic disease on resource utilization in common acute pediatric conditions. Financial concerns for children's hospitals.

OBJECTIVES: To estimate the resource utilization in hospitalizations for common pediatric conditions or procedures involving patients with chronic disease vs those with no chronic disease and to develop an economic model of hospital per-patient profit (or loss) when insurance contracts fail to account for the presence of chronic disease. SETTING AND DESIGN: A retrospective analysis of selected acute pediatric conditions found in the 1991 and 1992 MedisGroups National Comparative Data Base. PATIENTS: We studied 30379 pediatric admissions for common acute conditions, including concussion, croup, pneumonia, appendicitis, gastroenteritis, fractures, cellulitis, urinary tract infection, and viral illness. MAIN OUTCOME MEASURES: Hospital length of stay and total hospital charges. RESULTS: For patients without chronic disease, mean (geometric) length of stay was 2.53 vs. 3.05 days (P<.001) for patients with at least 1 chronic disease. For patients without chronic disease, mean (arithmetic) total hospital charge was S2614 vs. $3663 (P<.001) for patients with at least 1 chronic disease. Assuming 75% of patients with chronic disease are admitted to a children's hospital vs 25% to a general hospital, overall loss per patient at the children's hospital ranged between 1.5% and 2.9%, depending on assumptions regarding cost-to-charge ratios and the treatment of charge outliers. Pneumonia cases were associated with a 4.0% to 5.85% loss. CONCLUSIONS: Length of stay and charges are higher for everyday pediatric conditions or procedures when patients also have a chronic disease. If insurance contracts fail to account for chronic disease, then children's hospitals will realize significant financial losses, and over time this will lead to a decline in their financial viability, a reduction in quality, or a change in their mission.

First-cycle blood counts and subsequent neutropenia, dose reduction, or delay in early-stage breast cancer therapy.

PURPOSE: If patients could be ranked according to their projected need for supportive care therapy, then more efficient and less costly treatment algorithms might be developed. This work reports on the construction of a model of neutropenia, dose reduction, or delay that rank-orders patients according to their need for costly supportive care such as granulocyte growth factors. PATIENTS AND METHODS: A case series and consecutive sample of patients treated for breast cancer were studied. Patients had received standard-dose adjuvant chemotherapy for early-stage nonmetastatic breast cancer and were treated by four medical oncologists. Using 95 patients and validated with 80 additional patients, development models were constructed to predict one or more of the following events: neutropenia (absolute neutrophil count [ANC] < or = 250/microL), dose reduction > or = 15% of that scheduled, or treatment delay > or = 7 days. Two approaches to modeling were attempted. The pretreatment approach used only pretreatment predictors such as chemotherapy regimen and radiation history; the conditional approach included, in addition, blood count information obtained in the first cycle of treatment. RESULTS: The pretreatment model was unsuccessful at predicting neutropenia, dose reduction, or delay (c-statistic = 0.63). Conditional models were good predictors of subsequent events after cycle 1 (c-statistic = 0.87 and 0.78 for development and validation samples, respectively). The depth of the first-cycle ANC was an excellent predictor of events in subsequent cycles (P = .0001 to .004). Chemotherapy plus radiation also increased the risk of subsequent events (P = .0011 to .0901). Decline in hemoglobin (HGB) level during the first cycle of therapy was a significant predictor of events in the development study (P = .0074 and .0015), and although the trend was similar in the validation study, HGB decline failed to reach statistical significance. CONCLUSION: It is possible to rank patients according to their need of supportive care based on blood counts observed in the first cycle of therapy. Such rankings may aid in the choice of appropriate supportive care for patients with early-stage breast cancer.

Modeling the cost-effectiveness of granulocyte colony-stimulating factor use in early-stage breast cancer.

PURPOSE: To model the cost-effectiveness (CE) of granulocyte colony-stimulating factor (G-CSF) in early-stage breast cancer when its use is directed to those most in need of the medication. METHODS: A conditional CE model was developed for the use of G-CSF based on a ranking of patient need as determined by patient blood counts during the first cycle of chemotherapy. In the base case, no G-CSF was used. In the alternative case, G-CSF was used in the following manner. If the risk of a neutropenic event (as defined by a predictive model based on nadir absolute neutrophil count [ANC] and hemoglobin decrease in cycle 1) was equal to or exceeded a predetermined critical value "T," then patients would receive G-CSF in cycles 2 through 6 of chemotherapy. If the risk of an event was less than T, patients would not use G-CSF unless an event occurred, at which time G-CSF would be administered with every subsequent cycle. RESULTS: A decision rule (T) that would allow the most needy 50% of early-stage breast cancer patients to receive G-CSF after the first cycle of chemotherapy resulted in a CE ratio of $34,297 dollars per life-year saved (LYS). If only the most needy 10% of patients received G-CSF, then the associated CE ratio was $23,748/LYS; if 90% of patients could receive the medication, the CE ratio would be $76,487/LYS. These estimates were relatively insensitive to inpatient hospital cost estimates (inpatient costs for fever and neutropenia of $3,090 to $7,726 per admission produced dollar per LYS figures of $34,297 to $32,415, respectively). However, the model was sensitive to assumptions about the shape of the relationship between dose reduction and disease-free survival (DFS) at 3 years. CONCLUSION: Providing G-CSF to the neediest 50% of early-stage breast cancer patients (as defined by first-cycle blood counts) starting after the first cycle of chemotherapy is associated with a CE ratio of $34,297/LYS, which is well in the range of CE ratios for treatment of other common medical conditions. Furthermore, conditional CE studies, based on predictive models that incorporate individual patient risk, allow one to define populations for which therapy is, or is not, cost-effective. Limitations of our present understanding of the shape of the chemotherapy dose-response curve, especially at low levels of dose reductions, affect these results. Further work is required to define the shape of the dose-response curve in early-stage breast cancer.

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.

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.

Radiation-induced esophageal strictures in children with cancer.

The purpose of this study was to determine the long-term esophageal side effects of irradiation and (doxorubicin) chemotherapy given to children with cancer. Barium esophagograms and medical records of 18 patients with esophagitis who received between 1200 and 5580 cGy to the chest and chemotherapy were reviewed. The age range was 3-14 years. Esophageal strictures occurring 1-10 years after therapy were found in 5 patients with lymphoma; 3 of the five received doses of 4000 cGy or greater. Three children with esophagitis who received doses of 4000 cGy did not have stricture formation. However, their follow-up time was only 1-3 years. Of the 5 patients with esophageal stricture, 4 were treated with multiple dilatations, and a fifth required colonic interposition. Combined mediastinal irradiation and chemotherapy can result in esophageal stricture 1-10 years after therapy.

The effect of insurance status on likelihood of neonatal interhospital transfer.

OBJECTIVE: To determine the effect of insurance status on the likelihood of interhospital transfer for neonates. DESIGN: Population-based retrospective cohort study. SETTING: All general acute care nonpediatric hospitals in the five counties of southeastern Pennsylvania. PATIENTS: Fifty-six thousand, seven hundred eighty-nine infants from 0 to 28 days of age admitted to or born in study hospitals between January 1 and December 31, 1991. INTERVENTION: None. MAINS OUTCOME MEASURE: Transfer to another general or specialty acute care hospital. RESULTS: The incidence (95% confidence interval) of interhospital transfer was 1.69% (1.60, 1.78). Uninsured infants were nearly twice as likely [relative risk (RR) = 1.96 (1.67, 2.31)] to be transferred as commercially insured infants, even when adjusted for the effects of prematurity, severity of illness, and the level of neonatal intensive care unit in the referring hospital. Similarly, infants with Medicaid were more likely to be transferred [RR = 1.20 (1.01, 1.43)] than similar commercially insured neonates. Uninsured and publicly insured infants were also more likely to be born premature [RR 1.49 (1.39, 1. 60)] than privately insured neonates, and were more likely to have both moderate [RR 1.11 (1.04, 1.23)] and high [RR 1.21 (1.11, 1.32)] illness severity on admission to the hospital than privately insured infants. CONCLUSIONS: Neonates with no insurance and those with Medicaid coverage were more likely to be transferred than infants with private insurance. These results are consistent with those of other investigators who have studied financially motivated patient transfers- so-called patient dumping-in nonpediatric populations of patients. Our study may represent the first documentation of this phenomenon in a pediatric population. Our results are also consistent with those of other investigators who have examined the effect of insurance status on maternal interhospital transfer, thus providing further evidence for the existence of financially motivated transfers within regional systems of perinatal care. Future investigation into the effect of economic factors on variation in the utilization of transport services, and on how transfer influences ultimate patient outcome, is needed as managed care health systems become more widespread.

Successful management of low-stage neuroblastoma without adjuvant therapies: a comparison of two decades, 1972 through 1981 and 1982 through 1992, in a single institution.

PURPOSE: A review was undertaken of 119 children seen at the Children's Hospital of Philadelphia between 1972 and 1992 to assess the impact of adjuvant therapies for patients with low-stage neuroblastoma (NBL). PATIENTS AND METHODS: Twenty-one of 119 International Neuroblastoma Staging System (INSS) stage 1, 2a, 2b, and 4s patients seen received initial adjuvant treatment postoperatively and 98 did not. The patients were further subdivided according to decade, age, presence of residual disease, and lymph node status. Outcomes were then compared. RESULTS: The event-free survival (EFS) rate for those who received adjuvant therapy was 52% versus 86% for those who did not. The 5-year survival rate was 68% and 94%, respectively. Age (< or > 12 months), extent of residual disease, and status of lymph nodes did not influence survival. Over the two decades, the reasons for selecting treatment changed as new and powerful additional prognostic factors were identified; 71% of patients received no adjuvant treatment in the first decade, compared with 90% in the second. EFS rates for untreated patients by decade were 79% and 89%, and 5-year survival rates were 85% and 98%, respectively. CONCLUSION: It is possible to define most low-stage NBL as favorable-even in patients with positive lymph nodes and gross residual disease-and to omit initial adjuvant treatments successfully.