An introduction to multilevel regression models

Advancing current approaches to disease management evaluation: capitalizing on heterogeneity to understand what works and for whom

BACKGROUND

Evaluating large-scale disease management interventions implemented in actual health care settings is a complex undertaking for which universally accepted methods do not exist. Fundamental issues, such as a lack of control patients and limited generalizability, hamper the use of the 'gold-standard' randomized controlled trial, while methodological shortcomings restrict the value of observational designs. Advancing methods for disease management evaluation in practice is pivotal to learn more about the impact of population-wide approaches. Methods must account for the presence of heterogeneity in effects, which necessitates a more granular assessment of outcomes.

METHODS

This paper introduces multilevel regression methods as valuable techniques to evaluate 'real-world' disease management approaches in a manner that produces meaningful findings for everyday practice. In a worked example, these methods are applied to retrospectively gathered routine health care data covering a cohort of 105,056 diabetes patients who receive disease management for type 2 diabetes mellitus in the Netherlands. Multivariable, multilevel regression models are fitted to identify trends in clinical outcomes and correct for differences in characteristics of patients (age, disease duration, health status, diabetes complications, smoking status) and the intervention (measurement frequency and range, length of follow-up).

RESULTS

After a median one year follow-up, the Dutch disease management approach was associated with small average improvements in systolic blood pressure and low-density lipoprotein, while a slight deterioration occurred in glycated hemoglobin. Differential findings suggest that patients with poorly controlled diabetes tend to benefit most from disease management in terms of improved clinical measures. Additionally, a greater measurement frequency was associated with better outcomes, while longer length of follow-up was accompanied by less positive results.

CONCLUSIONS

Despite concerted efforts to adjust for potential sources of confounding and bias, there ultimately are limits to the validity and reliability of findings from uncontrolled research based on routine intervention data. While our findings are supported by previous randomized research in other settings, the trends in outcome measures presented here may have alternative explanations. Further practice-based research, perhaps using historical data to retrospectively construct a control group, is necessary to confirm results and learn more about the impact of population-wide disease management.

Multilevel Bayesian networks for the analysis of hierarchical health care data

OBJECTIVE

Large health care datasets normally have a hierarchical structure, in terms of levels, as the data have been obtained from different practices, hospitals, or regions. Multilevel regression is the technique commonly used to deal with such multilevel data. However, for the statistical analysis of interactions between entities from a domain, multilevel regression yields little to no insight. While Bayesian networks have proved to be useful for analysis of interactions, they do not have the capability to deal with hierarchical data. In this paper, we describe a new formalism, which we call multilevel Bayesian networks; its effectiveness for the analysis of hierarchically structured health care data is studied from the perspective of multimorbidity.

METHODS

Multilevel Bayesian networks are formally defined and applied to analyze clinical data from family practices in The Netherlands with the aim to predict interactions between heart failure and diabetes mellitus. We compare the results obtained with multilevel regression.

RESULTS

The results obtained by multilevel Bayesian networks closely resembled those obtained by multilevel regression. For both diseases, the area under the curve of the prediction model improved, and the net reclassification improvements were significantly positive. In addition, the models offered considerable more insight, through its internal structure, into the interactions between the diseases.

CONCLUSIONS

Multilevel Bayesian networks offer a suitable alternative to multilevel regression when analyzing hierarchical health care data. They provide more insight into the interactions between multiple diseases. Moreover, a multilevel Bayesian network model can be used for the prediction of the occurrence of multiple diseases, even when some of the predictors are unknown, which is typically the case in medicine.

Safety climate and medical errors in 62 US emergency departments

STUDY OBJECTIVE

We describe the incidence and types of medical errors in emergency departments (EDs) and assess the validity of a survey instrument that identifies systems factors contributing to errors in EDs.

METHODS

We conducted the National Emergency Department Safety Study in 62 urban EDs across 20 US states. We reviewed 9,821 medical records of ED patients with one of 3 conditions (myocardial infarction, asthma exacerbation, and joint dislocation) to evaluate medical errors. We also obtained surveys from 3,562 staff randomly selected from each ED; survey data were used to calculate average safety climate scores for each ED.

RESULTS

We identified 402 adverse events (incidence rate 4.1 per 100 patient visits; 95% confidence interval [CI] 3.7 to 4.5) and 532 near misses (incidence rate 5.4 per 100 patient visits; 95% CI 5.0 to 5.9). We judged 37% of the adverse events, and all of the near misses, to be preventable (errors); 33% of the near misses were intercepted. In multivariable models, better ED safety climate was not associated with fewer preventable adverse events (incidence rate ratio per 0.2-point increase in ED safety score 0.82; 95% CI 0.57 to 1.16) but was associated with more intercepted near misses (incidence rate ratio 1.79; 95% CI 1.06 to 3.03). We found no association between safety climate and violations of national treatment guidelines.

CONCLUSION

Among the 3 ED conditions studied, medical errors are relatively common, and one third of adverse events are preventable. Improved ED safety climate may increase the likelihood that near misses are intercepted.

Hospital-level associations with 30-day patient mortality after cardiac surgery: a tutorial on the application and interpretation of marginal and multilevel logistic regression

Background

Marginal and multilevel logistic regression methods can estimate associations between hospital-level factors and patient-level 30-day mortality outcomes after cardiac surgery. However, it is not widely understood how the interpretation of hospital-level effects differs between these methods.

Methods

The Australasian Society of Cardiac and Thoracic Surgeons (ASCTS) registry provided data on 32,354 patients undergoing cardiac surgery in 18 hospitals from 2001 to 2009. The logistic regression methods related 30-day mortality after surgery to hospital characteristics with concurrent adjustment for patient characteristics.

Results

Hospital-level mortality rates varied from 1.0% to 4.1% of patients. Ordinary, marginal and multilevel regression methods differed with regard to point estimates and conclusions on statistical significance for hospital-level risk factors; ordinary logistic regression giving inappropriately narrow confidence intervals. The median odds ratio, MOR, from the multilevel model was 1.2 whereas ORs for most patient-level characteristics were of greater magnitude suggesting that unexplained between-hospital variation was not as relevant as patient-level characteristics for understanding mortality rates. For hospital-level characteristics in the multilevel model, 80% interval ORs, IOR-80%, supplemented the usual ORs from the logistic regression. The IOR-80% was (0.8 to 1.8) for academic affiliation and (0.6 to 1.3) for the median annual number of cardiac surgery procedures. The width of these intervals reflected the unexplained variation between hospitals in mortality rates; the inclusion of one in each interval suggested an inability to add meaningfully to explaining variation in mortality rates.

Conclusions

Marginal and multilevel models take different approaches to account for correlation between patients within hospitals and they lead to different interpretations for hospital-level odds ratios.

Effects of the per diem prospective payment system with DRG-like grouping system (DPC/PDPS) on resource usage and healthcare quality in Japan

OBJECTIVES

In 2003, Japan introduced the prospective payment system (PPS) with diagnosis-related groups (DRG) rearranged grouping system called the diagnostic procedure combination/per-diem payment system (DPC/PDPS). Even after eight years, little is known about the effects of DPC/PDPS. The purpose of this study was to examine the effects of DPC/PDPS on resource usage and healthcare quality.

METHODS

Using 2001-2009 (fiscal year) administrative data of acute myocardial infarction patients, four indices, including inpatient total accumulated medical charges, length of stay (LOS), mortality rate, and readmission rate, were compared between patients reimbursed by DPC/PDPS or by fee-for-service.

RESULTS

DPC/PDPS significantly reduced total accumulated medical charges by $1061 (95% confidence interval [CI], -2007, -116) and LOS by 2.29 days (95% CI, -3.71, -0.88) after risk adjustment. However, mortality rate (Odds ratio [OR], 0.94; 95% CI, 0.73, 1.21) was unchanged. Furthermore, DPC/PDPS increased the readmission rate (OR, 1.37; 95% CI, 1.03, 1.82).

CONCLUSIONS

This study showed that DPC/PDPS was associated with reduced resource usage, but not improved healthcare quality, as with DRG/PPSs in other countries. To achieve successful healthcare reform, further discussion on additional motives will be required.

Administrative database research has unique characteristics that can risk biased results

OBJECTIVE

The provision of health care frequently creates digitized data--such as physician service claims, medication prescription records, and hospitalization abstracts--that can be used to conduct studies termed "administrative database research." While most guidelines for assessing the validity of observational studies apply to administrative database research, the unique data source and analytical opportunities for these studies create risks that can make them uninterpretable or bias their results.

STUDY DESIGN

Nonsystematic review.

RESULTS

The risks of uninterpretable or biased results can be minimized by; providing a robust description of the data tables used, focusing on both why and how they were created; measuring and reporting the accuracy of diagnostic and procedural codes used; distinguishing between clinical significance and statistical significance; properly accounting for any time-dependent nature of variables; and analyzing clustered data properly to explore its influence on study outcomes.

CONCLUSION

This article reviewed these five issues as they pertain to administrative database research to help maximize the utility of these studies for both readers and writers.

Influenza vaccine uptake among community-dwelling Italian elderly: results from a large cross-sectional study

Background

Flu vaccination significantly reduces the risk of serious complications like hospitalization and death among community-dwelling older people, therefore vaccination programmes targeting this population group represent a common policy in developed Countries. Among the determinants of vaccine uptake in older age, a growing literature suggests that social relations can play a major role.

Methods

Drawing on the socio-behavioral model of Andersen-Newman - which distinguishes predictors of health care use in predisposing characteristics, enabling resources and need factors - we analyzed through multilevel regressions the determinants of influenza immunization in a sample of 25,183 elderly reached by a nationally representative Italian survey.

Results

Being over 85-year old (OR = 1.99; 95% CI 1.77 - 2.21) and suffering from a severe chronic disease (OR = 2.06; 95% CI 1.90 - 2.24) are the strongest determinants of vaccine uptake. Being unmarried (OR = 0.81; 95% CI 0.74 - 0.87) and living in larger households (OR = 0.83; 95% CI 0.74 - 0.87) are risk factors for lower immunization rates. Conversely, relying on neighbors' support (OR = 1.09; 95% CI 1.02 - 1.16) or on privately paid home help (OR = 1.19; 95% CI 1.08 - 1.30) is associated with a higher likelihood of vaccine uptake.

Conclusions

Even after adjusting for socio-demographic characteristics and need factors, social support, measured as the availability of assistance from partners, neighbors and home helpers, significantly increases the odds of influenza vaccine use among older Italians.

Defining the Optimal Degree of Heparin Anticoagulation for Peripheral Vascular Interventions

Background—

The optimal degree of heparin anticoagulation for peripheral vascular interventions (PVIs) has not been defined. We sought to correlate total heparin dose and peak procedural activated clotting time (ACT) with postprocedural outcomes in patients undergoing PVI.

Methods and Results—

We studied 4743 patients who received heparin during PVIs in a regional, multicenter registry. From those, 1246 had recorded peak procedural ACT with the same point-of-care device. Periprocedural and in-hospital outcomes were compared between patients who received a total heparin dose <60 U/kg (n=2161) and ≥60 U/kg (n=2582). Similarly, outcomes were evaluated between groups with a peak procedural ACT <250 seconds (n=855) and ≥250 seconds (n=391). Technical and procedural success as well as intraprocedural thrombotic events did not differ between groups. Patients with heparin dose ≥60 U/kg had a higher rate of postprocedural hemoglobin drop ≥3 g/dL (7.09% versus 5.09%, respectively, P =0.004) and a higher transfusion rate compared with those with heparin dose <60 U/kg (4.92% versus 3.15%, respectively, P =0.002). In multivariate analysis, independent predictors of bleeding requiring transfusion were total heparin dose ≥60 U/kg, ACT ≥250 seconds, female sex, age ≥70 years, prior anemia, prior heart failure, low creatinine clearance, hybrid vascular surgery, rest pain, and below-knee intervention. In propensity-matched, risk-adjusted models and after hierarchical modeling, total heparin dose ≥60 U/kg and ACT ≥250 seconds remained strong predictors of post-PVI drop in hemoglobin ≥3 g/dL or transfusion.

Conclusions—

During PVI, higher total heparin dose (≥60 U/kg) and peak ACT ≥250 seconds were predictors of postprocedural transfusion. The high technical and procedural success in all groups suggests that use of weight-based heparin dosing with a target ACT <250 seconds in PVI may minimize the bleeding risk without compromising procedural success or increasing thromboembolic complications.

Unexplained international differences in clinical outcomes after acute myocardial infarction and fibrinolytic therapy: lessons from the Hirulog and Early Reperfusion or Occlusion (HERO)-2 trial

BACKGROUND

Despite advances in therapy, global mortality due to acute myocardial infarction remains high. The international Hirulog and Early Reperfusion or Occlusion (HERO-2) trial of 17,073 patients with ST-segment elevation myocardial infarction provided the opportunity to explore international differences in outcomes.

METHODS

Patient characteristics, treatment, and outcomes were compared across 5 diverse regions: Western countries, Latin America, Eastern Europe, Russia, and Asia. In addition, a representative sample of 1,743 screened patients was compared with enrolled patients.

RESULTS

Larger percentages of eligible patients were randomized in Eastern Europe, Russia, and Asia than Western countries. These regions enrolled more patients with anterior myocardial infarction, Killip class III or IV, and late presentation (>4 hours). More patients aged >75 years were enrolled from Western countries. Overall risk levels were similar. Eastern Europe and Russia had lower rates than Western countries of coronary revascularization (2% vs 18%) and longer hospital stays (median 18 vs 7 days). Thirty-day mortality was lower in Western countries; 6.7% versus 10.2% to 13.2% elsewhere, whereas reinfarction was more frequent (3.2% vs 1.5% to 3.0%; each, P < .001). Regional mortality differences persisted after adjustment for baseline risk factors, treatments, or national health and economic statistics (each P < .001).

CONCLUSIONS

The variation in mortality and other clinical outcomes across geographic regions was not adequately explained by risk factors, patterns of care, or national health statistics. Nevertheless, large international trials are a better way to assess potential new treatments across many countries than the alternative of separate smaller trials in each region.

Pre- and postnatal women's leisure time physical activity patterns: a multilevel longitudinal analysis

The purpose of this study was to examine women's leisure time physical activity (LTPA) before pregnancy, during pregnancy, and through the first 7 months postnatal. Pre- and postnatal women (n = 309) completed the 12-month Modifiable Activity Questionnaire and demographic information. Multilevel modeling was used to estimate a growth curve representing the average change in LTPA over time and intraindividual variations in the average growth curve over time. Growth curve estimates for the linear quadratic, and cubic trends were significant (p < .05), indicating that LTPA declined during pregnancy but then increased following birth. The results also demonstrated that the individual trajectories of LTPA varied substantially from the average growth curve. One demographic predictor variable (having other children at home) was significant (p < .05).

Identifying infants at increased risk for late initiation of immunizations: maternal and provider characteristics

OBJECTIVE

We identified maternal, provider, and community predictors among infants for late initiation of immunizations.

METHODS

We performed a retrospective cohort study of infants born between January 1, 2002, and December 31, 2004, in Philadelphia, Pennsylvania. Primary outcomes were age in days at first office-based immunization and status as a late starter (i.e., initiating office-based immunizations after 90 days of age). Candidate predictors included sociodemographic and prenatal characteristics, immunization provider practice type and size, and neighborhood factors. We performed hierarchical logistic regression and Cox regression models to identify independent predictors for being a late starter and prolonged time to first immunization.

RESULTS

Of the 65,519 infants from this birth cohort in Philadelphia's immunization registry, 54,429 (88.1%) were included in analysis and 12.6% of these were late starters. Infants whose mothers were younger, received less than five prenatal visits, had less than a high school education, had more than two children, and who smoked cigarettes prenatally were significantly more likely to be late starters. Receiving care at hospital/university-based or public health clinics was also significantly associated with likelihood of being a late starter. Neither distance between infant's residence and practice nor neighborhood socioeconomic indicators was independently associated with the outcomes. Common risk factor profiles based on practice type and four maternal characteristics were found to reliably identify infant risk.

CONCLUSIONS

Maternal receipt of fewer prenatal care visits, younger maternal age, higher birth order, and receiving care at public health clinics were the strongest predictors of being a late starter and time to first immunization. Risk factor profiles based on information already collected at birth can be used to identify higher-risk infants. Early intervention and potentially partnering with prenatal care providers may be key strategies for preventing underimmunization.

Are there regional differences in gynecologic cancer outcomes in the context of a single-payer, publicly-funded health care system? A population-based study

BACKGROUND

Canada has a single-payer, publicly-funded health care system that provides comprehensive health care, and therefore significant disparities in health outcomes are not expected in our population. The objective of this study was to determine if differences exist in endometrial cancer outcomes across regions in Ontario.

METHODS

This was a population-based study of all endometrial (uterine) cancer cases diagnosed from 1996 to 2000 in Ontario and linked to various administrative databases. Univariate analyses examined trends in demographics (age, income, co-morbidities), treatment (surgical staging and adjuvant pelvic radiotherapy), and pathology (grade, histology, stage) across 14 geographic regions defined by local health integration networks (LHINs) in Ontario. Primary outcome was 5-year overall survival among LHINs, which were compared in a multilevel Cox regression model to account for clustering of patient data at the hospital level.

RESULTS

There were 3,875 evaluable cases with complete information on demographics, treatment, pathology, and outcomes. There was significant variation in patient demographics, treatment, and pathology across the 14 LHINs. Low income level and surgery at a low-volume, community hospital without gynecologic oncologists were not associated with a higher risk of death. There was a trend towards clustering of patients within hospitals. After adjustment for covariates, there was no significant difference in survival across LHINs.

CONCLUSIONS

In the context of a single-payer, publicly-funded health care system, we did not find significant regional differences in endometrial cancer outcomes.

The Ontario printed educational message (OPEM) trial to narrow the evidence-practice gap with respect to prescribing practices of general and family physicians: a cluster randomized controlled trial, targeting the care of individuals with diabetes and hypertension in Ontario, Canada

Background

There are gaps between what family practitioners do in clinical practice and the evidence-based ideal. The most commonly used strategy to narrow these gaps is the printed educational message (PEM); however, the attributes of successful printed educational messages and their overall effectiveness in changing physician practice are not clear. The current endeavor aims to determine whether such messages change prescribing quality in primary care practice, and whether these effects differ with the format of the message.

Methods/design

The design is a large, simple, factorial, unblinded cluster-randomized controlled trial. PEMs will be distributed with informed, a quarterly evidence-based synopsis of current clinical information produced by the Institute for Clinical Evaluative Sciences, Toronto, Canada, and will be sent to all eligible general and family practitioners in Ontario. There will be three replicates of the trial, with three different educational messages, each aimed at narrowing a specific evidence-practice gap as follows: 1) angiotensin-converting enzyme inhibitors, hypertension treatment, and cholesterol lowering agents for diabetes; 2) retinal screening for diabetes; and 3) diuretics for hypertension.

For each of the three replicates there will be three intervention groups. The first group will receive informed with an attached postcard-sized, short, directive "outsert." The second intervention group will receive informed with a two-page explanatory "insert" on the same topic. The third intervention group will receive informed, with both the above-mentioned outsert and insert. The control group will receive informed only, without either an outsert or insert.

Routinely collected physician billing, prescription, and hospital data found in Ontario's administrative databases will be used to monitor pre-defined prescribing changes relevant and specific to each replicate, following delivery of the educational messages. Multi-level modeling will be used to study patterns in physician-prescribing quality over four quarters, before and after each of the three interventions. Subgroup analyses will be performed to assess the association between the characteristics of the physician's place of practice and target behaviours.

A further analysis of the immediate and delayed impacts of the PEMs will be performed using time-series analysis and interventional, auto-regressive, integrated moving average modeling.

Trial registration number

Current controlled trial ISRCTN72772651.

Effects of isometric handgrip training among people medicated for hypertension: a multilevel analysis

OBJECTIVE

To examine the longitudinal effects of isometric handgrip (IHG) exercise training on blood pressure using hierarchical linear modeling.

METHODS

Data from 43 participants who were medicated for hypertension at the time of training were amalgamated from three previous investigations. In each study, IHG training was completed 3 days/week for 8 weeks at 30% of maximal voluntary contraction and resting blood pressure was assessed at twice-weekly intervals throughout.

RESULTS

Hierarchical linear modeling analysis revealed a linear pattern of blood pressure decline over time with estimated reductions of 5.7 and 3 mmHg reductions in systolic and diastolic pressure, respectively. Participants with higher initial systolic pressure showed greater rates of blood pressure decline (r=-0.67), inferring that individuals with higher blood pressure stand to achieve greater benefits from this method of training.

CONCLUSIONS

These results provide further evidence that IHG training lowers resting blood pressure among persons medicated for hypertension.

Antibiotic treatment of wheezing in children with asthma: what is the practice?

OBJECTIVE

Antibiotics are not recommended for the treatment of wheezing in children with asthma, but little is known about their use. This study was undertaken to evaluate trends and determinants of antibiotic use in children with wheezing during the fiscal years 1995 through 2001.

METHODS

Using the population-based health care and prescription databases in Manitoba, Canada, this descriptive study examined time trends in antibiotic prescription use for wheezing episodes in a population of children with asthma. The likelihood of receiving an antibiotic prescription according to child and physician characteristics also was determined. Annual population-based rates of antibiotic prescriptions for wheezing episodes were modeled by age and antibiotic class, using general estimating equations. The odds ratio for receiving an antibiotic prescription according to child demographics and physician factors was determined from hierarchical linear modeling.

RESULTS

The antibiotic prescription rate for wheezing decreased by 28% from 708 prescriptions per 1000 children with asthma in 1995 to 511 prescriptions in 2001. Fifteen-fold increases in use were observed for broader spectrum macrolides in preschool children. Twenty-three percent of physician visits for wheezing resulted in an immediate antibiotic prescription, but this percentage increased to 64% for antibiotics that were received within 7 days of the episode. General practitioners prescribed antibiotics more often than did pediatricians. Physicians who were not trained in Canada or the United States were 40% more likely to prescribe antibiotics than their counterparts.

CONCLUSIONS

Antibiotic use for wheezing in children declined in the 1990s, but the increased use of broader spectrum macrolides has implications for antibiotic resistance. A link between antibiotic prescribing and physician specialty and location of training identifies opportunities for intervention.

Multilevel modeling was a convenient alternative to common regression designs in longitudinal suicide research

BACKGROUND AND OBJECTIVE

The prospective investigation of repetitive nonfatal suicidal behavior is associated with two methodological problems. Due to the commonly used definitions of nonfatal suicidal behavior, clinical samples usually consist of patients with a considerable between-person variability. Second, repeated nonfatal suicidal episodes of the same subjects are likely to be correlated. We examined three regression techniques to comparatively evaluate their efficiency in addressing the given methodological problems.

STUDY DESIGN AND SETTING

Repeated episodes of nonfatal suicidal behavior were assessed in two independent patient samples during a 2-year follow-up period. The first regression design modeled repetitive nonfatal suicidal behavior as a summary measure. The second regression model treated repeated episodes of the same subject as independent events. The third regression model represented a hierarchical linear model.

RESULTS

The estimated mean effects of the first model were likely to be nonrepresentative for a considerable part of the study subjects. The second regression design overemphasized the impact of the predictor variables. The hierarchical linear model most appropriately accounted for the heterogeneity of the samples and the correlated data structure.

CONCLUSION

The nonhierarchical regression designs did not provide appropriate statistical models for the prospective investigation of repetitive nonfatal suicidal behavior. Multilevel modeling provides a convenient alternative.

Regional variation in rates of low birth weight

OBJECTIVE

Low birth weight (LBW; < 2500 g) is the result of complex and poorly understood interactions between the biological determinants of the mother and the fetus, the parent's socioeconomic status, and medical care. After controlling for these established risk factors, the extent of regional variation in LBW rates remains unknown. This study measures regional variation in LBW rates and identifies regions of neonatal health services with significantly high or low adjusted rates.

METHODS

Linking the United States 1998 singleton birth cohort (N = 3.8 million) with county and health care characteristics, we conducted a small area analysis of LBW across 246 regions of neonatal health services. We measured observed rates and then used a multivariable, hierarchical model to estimate adjusted LBW rates by regions. We then stratified these rates by race for the 208 regions with adequate sample size.

RESULTS

Observed LBW rates varied across regions from 3.8 to 10.6 per 100 live births (interquartile range: 5.0-6.8 [25th-75th percentile]; median: 5.9). After controlling for known maternal and area risk factors, 67 (27.0%) regions had rates significantly below and 98 (39.8%) regions had rates significantly higher than the national rate of 6.0 per 100 live births. Although black mothers were more likely to give birth to an LBW newborn, regional adjusted rates still varied > 3-fold within both black and nonblack subgroups.

CONCLUSIONS

After controlling for known maternal and area risk factors, LBW rates markedly varied across US regions of neonatal health services for both black and nonblack mothers. Additional analyses of these regions may provide opportunities for regional accountability in pregnancy outcomes, LBW research, and targeted improvement interventions, especially in high-risk populations.

Clinical and Biologic Factors Related to Oral Implant Failure: A 2-Year Follow-Up Study

The purpose of this study is to evaluate urinary biomarkers of bone formation and resorption as predictive factors for oral implant failure, and to contribute to the knowledge of factors related to oral implant failure. A total of 93 patients between 18 and 85 years old, with an indication of oral implant, were eligible in this 2-year prospective, open, and nonrandomized study. Patients who had bone graft before implantation or presented with prosthetic difficulties (implant-to-crown ratio < 1, and/or unfavorable intermaxillary space) were excluded. All patients received either Frialit-2 (Friadent, Mannheim, Germany), cylinder, or screwed implants or IMZ Twin Plus (Friadent), cylinder implants, with FRIOS (Friadent) titanium coating. Serum osteocalcin, and urinary pyridinoline and deoxypyridinoline were measured, together with bone density at implant location. The primary endpoint (implant failure) was the implant removal (radiographic evidence of peri-implant bone loss and/or pockets). Factors related to implant failure were analyzed using multilevel logistic regression models to consider within-patient effects. Of the 93 patients included, 61% were female, and 16% were current smokers. A total of 266 oral implants were placed and analyzed, with a mean number of 3.1 implants by patient. Eleven and 15% of bone locations scored at D1 and D4, respectively, for the Misch bone density scoring. The majority of implants (72%) were placed more than 3 months after tooth extraction, using a Frialit-2 system in 73% of cases. The mean of osteocalcin was 17.3 (+/-9.4) ng/L; those of pyridinoline and deoxypyridinoline were 33.2 (+/-15.8) and 10.2 (+/-11.9) mmol per creatinine mmol, respectively. At one-year, 95.5% (95% confidence interval 92.5-97.5) of implants have not been removed. One year later, no further implant failed. In both univariate and multivariate analysis, osteocalcin, pyridinoline, and deoxypyridinoline were not significant predictive factors of oral implant failure. In multilevel logistic regression analysis, only tobacco consumption and single-tooth replacement or removable prosthesis were independent and significant predictive factors of oral implant failure. Serum osteocalcin, and urinary pyridinoline and deoxypyridinoline were not predictive of oral implant failure in this study. These results suggest that oral implants are more likely to fail for posterior single-tooth replacements and removable prostheses rather than for complete edentulous fixed bridgeworks or overdentures. Tobacco smoking has been identified as a major risk factor of oral implant failure.

Estimating the annual rate of de novo multiple aneurysms: three statistical approaches

Object

Individuals with unruptured intracranial aneurysms experience a higher rate of rupture if their history includes another aneurysm that has previously bled. The authors used systematic review and metaregression to estimate the annual rate of development of second de novo aneurysms after subarachnoid hemorrhage.

Methods

This investigation included studies in which more than 300 patients with intracranial aneurysms were described, and in which the age of the patients and the proportion with multiple aneurysms were documented. Studies describing delayed follow-up angiography that was performed after treatment of aneurysms were also reviewed.

Twenty studies were included in a between-study analysis. The univariate odds ratio (OR) for multiple intracranial aneurysms per year of age was 1.085 (95% confidence interval [CI] 1.015–1.165); this value was calculated using a hierarchical model for between-study heterogeneity. Five studies were included that provided age stratification. The estimated OR for multiple intracranial aneurysms per year was 1.011 (95% CI 1.005–1.018). Four follow-up studies were available.

Conclusions

According to the three different approaches (study-level, patient-level, and follow-up analyses), the estimated annual rates of development of de novo aneurysms were 1.62% (95% CI 0.28–3.59%), 0.28% (95% CI 0.12–0.49%), and 0.92% (95% CI 0.64–1.25%), respectively. The estimated annual rate of development of second de novo aneurysms ranged from 0.28 to 1.62%.

Measuring health and disease: progress in analytical approaches

In this paper, Dr. Martin reviews the progress in analytical approaches used in veterinary medicine between the 1970s and today. The newer applications are used in such activities as monitoring/surveillance, analysis of observational study data, evaluation of tests in the absence of gold standards, the analysis of clustered data (including geographically clustered data) and modeling disease in populations. Future work will be more complex but will demand an increased emphasis on ways to enhance our biological understanding of the results of data analysis and modeling.

Individual and neighbourhood determinants of health care utilization. Implications for health policy and resource allocation

OBJECTIVES

To investigate the importance of both individual and neighbourhood socioeconomic characteristics for health care utilization.

METHODS

Various linkage procedures generated a longitudinal dataset with information on 2,116 Nova Scotians, their residential neighbourhoods, 8 years of health care utilization and vital status. Unilevel and multilevel regression analyses were employed to examine the effects of both individual and neighbourhood characteristics on health care use.

RESULTS

Individual income and education determined physician and hospital use. Also, neighbourhood characteristics, specifically average income and percentage of single mother families, were found to determine health care use. When considering individual and neighbourhood characteristics simultaneously, individual income and education determined physician and hospital use independently, while neighbourhood income determined physician use independently.

CONCLUSIONS

Both individual and neighbourhood socioeconomic characteristics determine health care use. Acknowledging this allows better targeting of health policy and planning, and enables more accurate needs-based resource allocation.

Individual and neighbourhood determinants of health care utilization. Implications for health policy and resource allocation

OBJECTIVES

To investigate the importance of both individual and neighbourhood socioeconomic characteristics for health care utilization.

METHODS

Various linkage procedures generated a longitudinal dataset with information on 2,116 Nova Scotians, their residential neighbourhoods, 8 years of health care utilization and vital status. Unilevel and multilevel regression analyses were employed to examine the effects of both individual and neighbourhood characteristics on health care use.

RESULTS

Individual income and education determined physician and hospital use. Also, neighbourhood characteristics, specifically average income and percentage of single mother families, were found to determine health care use. When considering individual and neighbourhood characteristics simultaneously, individual income and education determined physician and hospital use independently, while neighbourhood income determined physician use independently.

CONCLUSIONS

Both individual and neighbourhood socioeconomic characteristics determine health care use. Acknowledging this allows better targeting of health policy and planning, and enables more accurate needs-based resource allocation.