Comparing hierarchical modeling with traditional logistic regression analysis among patients hospitalized with acute myocardial infarction: Should we be analyzing cardiovascular outcomes data differently?

doi:10.1067/mhj.2003.23

American Heart Journal

Volume 145, Issue 1, January 2003, Pages 27-35

https://doi.org/10.1067/mhj.2003.23 Get rights and content

Abstract

Background Data in health research are frequently structured hierarchically. For example, data may consist of patients treated by physicians who in turn practice in hospitals. Traditional statistical techniques ignore the possible correlation of outcomes within a given practice or hospital. Furthermore, imputing characteristics measured at higher levels of the hierarchy to the patient-level artificially inflates the amount of available information on the effect of higher-level characteristics on outcomes.Methods Conventional logistic regression models and multilevel logistic regression models were fit to a cross-sectional cohort of patients hospitalized with a diagnosis of acute myocardial infarction. The statistical significance of the effect of patient, physician, and hospital characteristics on patient outcomes was compared between the 2 modeling strategies.Results The 2 analytic strategies agreed well on the effect of patient characteristics on outcomes. According to the traditional analysis, teaching status was statistically significantly associated with 5 of the 9 outcomes, whereas the multilevel models did not find a statistically significant association between teaching status and any patient outcomes. Similarly, the traditional and multilevel models disagreed on the statistical significance of the effect of being treated at a revascularization hospital and 3 patient outcomes.Conclusions In comparing the resultant models, we see that false inferences can be drawn by ignoring the structure of the data. Conventional logistic regression tended to increase the statistical significance for the effects of variables measured at the hospital-level compared to the level of significance indicated by the multilevel model. (Am Heart J 2003;145:27-35.)

References (0)

Cited by (122)

Out-of-hospital cardiac arrest survival when CPR is initiated by first responders
2023, Resuscitation
In most patients with out-of-hospital cardiac arrest (OHCA), cardiopulmonary resuscitation (CPR) is initiated by first responders (non-transporting firefighters or police) or emergency medical service (EMS) personnel. Whether survival outcomes differ when CPR is initiated by first responders vs. EMS is unclear.
Within the CARES registry, we identified 162,896 adult patients with a non-traumatic OHCA in whom CPR was initiated by first responders or EMS during 2013–2021. Using multivariable hierarchical logistic regression to adjust for demographics, cardiac arrest characteristics and time to first CPR, we compared rates of survival to hospital admission and to discharge in patients with CPR initiated by first responders and EMS.
CPR was initiated by first responders in 70,889 (43.5%) and by EMS in 92,007 (56.5%) patients. Time to first CPR was shorter when first responders initiated CPR (median: 8.0 [5.0–13.0] vs. 10.0 minutes [IQR: 6.0–14.0]; standardized difference 16.1%). The likelihood of survival to hospital admission was similar when CPR was initiated by first responders (27.1% [first responders] vs. 26.8% [EMS]; adjusted OR: 0.98 [0.96, 1.01], P = 0.15) whereas survival rates to discharge were higher with CPR initiated by first responders (9.4% [first responders] vs. 7.7% [EMS]; adjusted OR: 1.17 [1.02, 1.21], P < 0.001). After adjustment for time to first CPR, rates of survival to discharge were similar between the first responder and EMS groups (adjusted OR: 1.04 [1.00–1.08]; P = 0.07).
CPR initiated by first responders for OHCA is associated with higher overall survival rates and higher survival was largely mediated by earlier response times.
Textbook outcome after rectal cancer surgery as a composite measure for quality of care: A population-based study
2021, European Journal of Surgical Oncology
Textbook outcome is a composite measure of combined outcome indicators, which has been suggested to be of additional value over single outcome parameters in clinical auditing of surgical treatment. This study aimed to assess textbook outcome after rectal cancer surgery as short-term marker for quality of care.
Patients who underwent elective rectal cancer surgery between 2012 and 2019 and registered in the Dutch ColoRectal Audit were included. Textbook outcome was achieved when the following criteria were met: 30-day and primary hospital admission survival, no reintervention, tumour-free margins, no postoperative complications, a hospital stay of less than 14 days and no readmission. Hospital variation was evaluated in case-mix corrected funnel-plots. A multilevel logistic regression analysis was performed to identify associated factors with textbook outcome.
The study population consisted of 20,521 patients who underwent primary rectal cancer surgery, of whom 56.3% achieved textbook outcome. Postoperative complications were the main contributor to not achieving textbook outcome. Case-mix corrected funnel plots demonstrated that underperforming hospitals in 2012–2015 were no underperformers in 2016–2019 anymore. Female sex, laparoscopic surgery, and rectal resection without defunctioning stoma creation were positively associated with textbook outcome.
Textbook outcome after rectal cancer resection is mainly driven by postoperative complications. Although textbook outcome showed some discriminating value for identifying underperforming hospitals, it does not fit the plan-do-check-act cycle of clinical auditing. In our opinion, textbook outcome has little added value to the current outcome indicators for rectal cancer surgery.
Optimal Hospital and Surgeon Volume Thresholds to Improve 30-Day Readmission Rates, Costs, and Length of Stay for Total Hip Replacement
2019, Journal of Arthroplasty
Little is known about whether there are optimal hospital and surgeon volume thresholds to reduce readmission, costs, and length of stay (LOS) for total hip replacement (THR). Nationwide population-based data were applied to identify the optimal hospital and surgeon volume thresholds and to discover the effects of these volume thresholds on 30-day unplanned readmission, costs and LOS for THR.
A total of 6367 patients identified through Taiwan’s National Health Insurance Research Database received THR in 2012. Restricted cubic splines were used to identify the optimal hospital and surgeon volume needed to decrease the risk of 30-day unplanned readmission. Multilevel regression modeling and propensity score weighting were used to examine the impact of hospital and surgeon volume thresholds on 30-day unplanned readmission, costs, and LOS, after adjusting for patient, surgeon, and hospital characteristics.
The volume thresholds for hospitals and surgeons were 65 cases and 15 cases a year, respectively. The overall mean LOS was 7.3 ± 4.3 days. Patients who received THR from surgeons who did not reach the volume threshold had higher 30-day unplanned readmission rates, costs, and LOS than those who received THR from surgeons who reached the volume threshold.
This is the first study to identify the surgeon volume threshold that can reduce 30-day unplanned readmission rates, costs, and LOS for THR. However, the results from Taiwan may not be applicable to other parts of the world. Identifying the threshold could help patients, providers, and policymakers to make decisions regarding optimal delivery of THR.
External benchmarking of trauma services in New South Wales: Risk-adjusted mortality after moderate to severe injury from 2012 to 2016
2019, Injury
Citation Excerpt :
The main outcome measure was in-hospital death. Given the nested structure of the data, hierarchical logistic regression models were used [14]. Patients were treated as the lower level units which were nested within each centre (higher level units).
Trauma centres and systems have been associated with improved morbidity and mortality after injury. However, variability in outcomes across centres within a given system have been demonstrated. Performance improvement initiatives, that utilize external benchmarking as the backbone, have demonstrated system-wide improvements in outcomes. This data driven approach has been lacking in Australia to date. Recent improvement in local data quality may provide the opportunity to engage in data driven performance improvement. Our objective was to generate risk-adjusted outcomes for the purpose of external benchmarking of trauma services in New South Wales (NSW) based on existing data standards.
Retrospective cohort study of the NSW Trauma Registry. We included adults (>16 years), with an Injury Severity Score >12, that received definitive care at either Major Trauma Services (MTS) or Regional Trauma Services (RTS) between 2012-2016. Hierarchical logistic regression models were then used to generate risk-adjusted outcomes. Our outcome measure was in-hospital death. Demographics, vital signs, transfer status, survival risk ratios, and injury characteristics were included as fixed-effects. Median odds ratios (MOR) and centre-specific odds ratios with 95% confidence intervals were generated. Centre-level variables were explored as sources of variability in outcomes.
14,452 patients received definitive care at one of seven MTS (n = 12,547) or ten RTS (n = 1905). Unadjusted mortality was lower at MTS (9.4%) compared to RTS (11.2%). After adjusting for case-mix, the MOR was 1.33, suggesting that the odds of death was 1.33-fold greater if a patient was admitted to a randomly selected centre with worse as opposed to better risk-adjusted mortality. Definitive care at an MTS was associated with a 41% lower likelihood of death compared to definitive care at an RTS (OR 0.59 95%CI 0.35-0.97). Similar findings were present in the elderly and isolated severe brain injury subgroups.
The NSW trauma system exhibited variability in risk-adjusted outcomes that did not appear to be explained by case-mix. A better understanding of the drivers of the described variation in outcomes is crucial to design targeted locally-relevant quality improvement interventions.
Accounting for clustering in automated variable selection using hospital data: a comparison of different LASSO approaches
2023, BMC Medical Research Methodology
Accounting for clustering in automated variable selection using hospital data: A comparison of different LASSO approaches
2023, Research Square

View all citing articles on Scopus

^☆: Supported in part by an operating grant from the Canadian Institutes of Health Research (CIHR). Dr Tu is supported by a Canada Research Chair in Health Services Research. The Institute for Clinical Evaluative Sciences is supported in part by a grant from the Ontario Ministry of Health and Long Term Care.

View full text

Clinical InvestigationsComparing hierarchical modeling with traditional logistic regression analysis among patients hospitalized with acute myocardial infarction: Should we be analyzing cardiovascular outcomes data differently?☆

Abstract

Clinical Investigations
Comparing hierarchical modeling with traditional logistic regression analysis among patients hospitalized with acute myocardial infarction: Should we be analyzing cardiovascular outcomes data differently?☆