Fitting mixed-effects models for repeated ordinal outcomes with the NLMIXED procedure

Predictors of residential mobility and its impact on air pollution exposure among children diagnosed with early childhood leukemia

Epidemiology studies relying on one address to assign exposures over time share common methodological limitations in failing to account for mobility that may introduce potential exposure misclassification. Using Texas birth certificate and cancer registry data, we identified predictors of residential mobility among mothers of children diagnosed with early childhood leukemia in Texas from 1995 to 2011. We used U.S. Environmental Protection Agency (EPA) National Air Toxics Assessment data to estimate residential levels of benzene and 1,3-butadiene based on addresses at birth and diagnosis and applied mixed-effects ordinal logistic regression models to evaluate differences in exposure classification between the two time periods. In total, 55% of children moved from time of birth to diagnosis, although they generally did not move far (median distance moved was 8 km). Predictors of mobility, at delivery, included younger age, being unmarried and living in neighborhoods with high benzene levels, and, at diagnosis, increasing child's age and living in neighborhoods with low poverty rates. We observed that the odds of being assigned to a higher exposure quartile at diagnosis relative to the time of birth decreased by 31% for 1,3-butadiene (OR = 0.69, 95% CI 0.59-0.82) and by 12% for benzene (OR = 0.88, 95% CI 0.75, 1.05).

Differences in environmental exposure assignment due to residential mobility among children with a central nervous system tumor: Texas, 1995-2009

In epidemiologic studies of childhood cancer, environmental exposures are often assigned based on either residence at birth or diagnosis without considering the impact of residential mobility. Therefore, we evaluated residential mobility and exposure assignment differences to hazardous air pollutants between birth and diagnosis in children with a central nervous system (CNS) tumor. Children diagnosed with CNS tumors during 1995-2009 (N=1,196) were identified from the Texas Cancer Registry. Census tract-level estimates of 1,3-butadiene and benzene were used to assign quartiles of exposure based on the maternal residence at birth and the child's residence at diagnosis. Overall, 64% of younger (0-4 years) children and 79% of older (5-14 years) children moved between birth and diagnosis. Using mixed-effects ordinal logistic regression, residence at diagnosis compared to birth did not result in a significant change in exposure assignment for younger children; however, older children were more likely to be placed in a lower 1,3-butadiene or benzene exposure quartile based on residence at diagnosis compared to birth (odds ratio (OR)=0.58, 95% confidence interval (CI)=0.45-0.76; OR=0.57, 95% CI=0.44-0.75, respectively). In conclusion, while the majority of children moved between birth and CNS tumor diagnosis, mobility did not significantly impact 1,3-butadiene and benzene exposure assessment in younger children.

Dream Recall and the Full Moon

There is ongoing debate on whether the full moon is associated with sleep and dreaming. The analysis of diaries kept by the participants ( N = 196) over 28 to 111 nights showed no association of a full moon and dream recall. Psychological factors might explain why some persons associate a full moon with increased dream recall.

Changes in sexual problems over time in women with and without early-stage breast cancer

OBJECTIVE

We aimed to evaluate whether age-matched women with and without early-stage breast cancer differentially experience sexual problems over time and whether changes in patients' problems differ by type of surgical procedure.

METHODS

We conducted four interviews (4-6 wk, 6 mo, and 1 and 2 y) after tumor resection (patients) or after a negative/benign screening mammogram (controls). Mixed-effects models with repeated ordinal measurements tested the effects of time and diagnostic group (stages 0, I, and IIA vs controls) and, for patients, the effects of time and type of surgical procedure (mastectomy vs lumpectomy) using a newly developed nine-item sexual-problem measure. Two-sided P values < 0.05 were considered significant.

RESULTS

Using data from 1,033 women (17.3% stage 0, 33.4% stage I/IIA, and 49.3% controls; mean age, 57.1 y; 23.1% nonwhite; 64.7% married), two factors measuring problems with sexual attractiveness and sexual interest/enjoyment emerged in exploratory factor analysis (alpha > or = 0.74 for each subscale and the nine-item measure). Patients and controls reported few sexual problems on average, but controls were more likely to report sexual problems on the nine-item measure over time and stage I patients were less likely to report problems with sexual attractiveness over time (each P < 0.05) compared with at baseline. Patients who underwent mastectomy (35.7%) were 2.7 times more likely to report sexual problems on the nine-item measure at 2-year follow-up compared with baseline (P = 0.0339).

CONCLUSIONS

Patients and controls experienced few sexual problems over time, and in fact, controls were more likely to report sexual problems at subsequent interviews, whereas patients were not.

Differences in the quality of interpersonal care in complementary and conventional medicine

BACKGROUND

The study was part of a nationwide evaluation of complementary and alternative medicine (CAM) in Swiss primary care. The aim of the study was to compare patient-physician relationships and the respective patient-reported relief of symptoms between CAM and conventional primary care (COM).

METHODS

A comparative observational study in Swiss primary care with written survey completed by patients who visited a GP one month earlier. 6133 patients older than 16 years of 170 certified CAM physicians, of 77 non-certified CAM physicians and of 71 conventional physicians were included. Patients completed a questionnaire aimed at symptom relief, patient satisfaction, fulfilment of expectations, and quality of patient-physician interaction (EUROPEP questionnaire).

RESULTS

CAM physicians treated significantly more patients with chronic conditions than COM physicians. CAM Patients had significant higher healing expectations than COM patients. General patient satisfaction was significantly higher in CAM patients, although patient-reported symptom relief was significantly poorer. The quality of patient-physician communication was rated significantly better in CAM patients.

CONCLUSIONS

The study shows better patient-reported outcomes of CAM in comparison to COM in Swiss primary care, which is related to higher patient satisfaction due to better patient-physician communication of CAM physicians. More effective communication patterns of these physicians may play an important role in allowing patients to maintain more positive outcome expectations. The findings should promote formative efforts in conventional primary care to improve communication skills in order to reach the same levels of favourable patient outcomes.

A longitudinal study of factors associated with perceived risk of recurrence in women with ductal carcinoma in situ and early-stage invasive breast cancer

Breast cancer patients' perceived risk of recurrence has been associated with psychological distress. Little is known about the change of patients' perceived risk of recurrence over time and factors associated with their recurrence-risk perceptions. We prospectively recruited 549 newly diagnosed early-stage breast cancer patients; patients completed interviews at 6 weeks, 6 months, 1 year, and 2 years after definitive surgical treatment. A random-effects regression model with repeated ordinal measurements was used to estimate the relationship between perceived risk of recurrence and demographic, medical, and psychosocial factors. We analyzed data from 535 patients [34% ductal carcinoma in situ (DCIS); 20% non-white] who reported their perceived risk at one or more interviews. At the first interview, 16% reported having no lifetime risk of recurrence, and another 16% reported ≥ 50% risk of recurrence, including 15% of DCIS patients. Patients who were white (OR = 5.88, 95% CI 3.39-10.19) and had greater state anxiety (OR = 1.04, 95% CI 1.02-1.07) were more likely, while patients who received radiotherapy (OR = 0.72, 95% CI 0.54-0.96) and had more social support (OR = 0.59, 95% CI 0.46-0.75) were less likely to report higher risk of recurrence. Cancer stage was not significantly associated with perceived risk of recurrence. Perceived risk of recurrence did not change significantly over time. Educating early-stage breast cancer patients about their actual risk could result in more realistic recurrence-risk perceptions, and increasing social support could help alleviate anxiety associated with exaggerated risk perceptions.

Differences in exposure assignment between conception and delivery: the impact of maternal mobility

In studies of reproductive outcomes, maternal residence at delivery is often the only information available to characterise environmental exposures during pregnancy. The goal of this investigation was to describe residential mobility during pregnancy and to assess the extent to which change of residence may result in exposure misclassification when exposure is based on the address at delivery. Maternal residential mobility was compared between neural tube defect cases and unaffected controls from Texas participants in the National Birth Defects Prevention Study (NBDPS). Maternal residential information was obtained from the NBDPS interview. Data from the U.S. EPA National Air Toxics Assessment [Assessment System for Population Exposure Nationwide (ASPEN)], modelled at the census tract level, were used to estimate benzene exposure based on address at conception and address at delivery. Quartiles of exposure were assigned based on these estimates and the quartile assignments based on address at conception and address at delivery were compared using traditional methods (kappa statistics) and a novel application of mixed-effects ordinal logistic regression. Overall, 30% of case mothers and 24% of control mothers moved during pregnancy. Differences in maternal residential mobility were not significant between cases and controls, other than case mothers who moved did so earlier during pregnancy than control mothers (P = 0.01). There was good agreement between quartiles of estimated benzene exposure at both addresses (kappa = 0.78, P < 0.0001). Based on the mixed-effects regression model, address at delivery was not significantly different from using address at conception when assigning quartile of benzene exposure based on estimates from ASPEN (odds ratio 1.03, 95% confidence interval 0.85, 1.25). Our results indicate that, in this Texas population, maternal residential movement is generally within short distances, is typically not different between cases and controls, and does not significantly influence benzene exposure assessment.

Professional judgment and the interpretation of viable mold air sampling data

Although mold air sampling is technically straightforward, interpreting the results to decide if there is an indoor source is not. Applying formal statistical tests to mold sampling data is an error-prone practice due to the extreme data variability. With neither established exposure limits nor useful statistical techniques, indoor air quality investigators often must rely on their professional judgment, but the lack of a consensus "decision strategy" incorporating explicit decision criteria requires professionals to establish their own personal set of criteria when interpreting air sampling data. This study examined the level of agreement among indoor air quality practitioners in their evaluation of airborne mold sampling data and explored differences in inter-evaluator assessments. Eighteen investigators independently judged 30 sets of viable mold air sampling results to indicate: "definite indoor mold source," "likely indoor mold source," "not enough information to decide," "likely no indoor mold source," or "definitely no indoor mold source." Kappa coefficient analysis indicated weak inter-observer reliability, and comparison of evaluator mean scores showed clear inter-evaluator differences in their overall scoring patterns. The responses were modeled on indicator "traits" of the data sets using a generalized, linear mixed model approach and showed several traits to be associated with respondents' ratings, but they also demonstrated distinct and divergent inter-evaluator response patterns. Conclusions were that there was only weak overall agreement in evaluation of the mold sampling data, that particular traits of the data were associated with the conclusions reached, and that there were substantial inter-evaluator differences that were likely due to differences in the personal decision criteria employed by the individual evaluators. The overall conclusion was that there is a need for additional work to rigorously explore the constellation of decision criteria, the weightings employed by individual practitioners, and the rationale under which criteria are adopted as first steps toward the larger goal of developing a consensus mold decision strategy.

Patient satisfaction with primary care: an observational study comparing anthroposophic and conventional care

Background

This study is part of a cross-sectional evaluation of complementary medicine providers in primary care in Switzerland. It compares patient satisfaction with anthroposophic medicine (AM) and conventional medicine (CON).

Methods

We collected baseline data on structural characteristics of the physicians and their practices and health status and demographics of the patients. Four weeks later patients assessed their satisfaction with the received treatment (five items, four point rating scale) and evaluated the praxis care (validated 23-item questionnaire, five point rating scale). 1946 adult patients of 71 CON and 32 AM primary care physicians participated.

Results

1. Baseline characteristics: AM patients were more likely female (75.6% vs. 59.0%, p < 0.001) and had higher education (38.6% vs. 24.7%, p < 0.001). They suffered more often from chronic illnesses (52.8% vs. 46.2%, p = 0.015) and cancer (7.4% vs. 1.1%). AM consultations lasted on average 23,3 minutes (CON: 16,8 minutes, p < 0.001).

2. Satisfaction: More AM patients expressed a general treatment satisfaction (56.1% vs. 43.4%, p < 0.001) and saw their expectations completely fulfilled at follow-up (38.7% vs. 32.6%, p < 0.001). AM patients reported significantly fewer adverse side effects (9.3% vs. 15.4%, p = 0.003), and more other positive effects from treatment (31.7% vs. 17.1%, p < 0.001).

Europep: AM patients appreciated that their physicians listened to them (80.0% vs. 67.1%, p < 0.001), spent more time (76.5% vs. 61.7%, p < 0.001), had more interest in their personal situation (74.6% vs. 60.3%, p < 0.001), involved them more in decisions about their medical care (67.8% vs. 58.4%, p = 0.022), and made it easy to tell the physician about their problems (71.6% vs. 62.9%, p = 0.023). AM patients gave significantly better rating as to information and support (in 3 of 4 items p [less than or equal to] 0.044) and for thoroughness (70.4% vs. 56.5%, p < 0.001).

Conclusion

AM patients were significantly more satisfied and rated their physicians as valuable partners in the treatment. This suggests that subject to certain limitations, AM therapy may be beneficial in primary care. To confirm this, more detailed qualitative studies would be necessary.

Patient satisfaction and side effects in primary care: an observational study comparing homeopathy and conventional medicine

Background

This study is part of a nationwide evaluation of complementary medicine in Switzerland (Programme Evaluation of Complementary Medicine PEK) and was funded by the Swiss Federal Office of Public Health. The main objective of this study is to investigate patient satisfaction and perception of side effects in homeopathy compared with conventional care in a primary care setting.

Methods

We examined data from two cross-sectional studies conducted in 2002–2003. The first study was a physician questionnaire assessing structural characteristics of practices. The second study was conducted on four given days during a 12-month period in 2002/2003 using a physician and patient questionnaire at consultation and a patient questionnaire mailed to the patient one month later (including Europep questionnaire).

The participating physicians were all trained and licensed in conventional medicine. An additional qualification was required for medical doctors providing homeopathy (membership in the Swiss association of homeopathic physicians SVHA).

Results

A total of 6778 adult patients received the questionnaire and 3126 responded (46.1%). Statistically significant differences were found with respect to health status (higher percentage of chronic and severe conditions in the homeopathic group), perception of side effects (higher percentage of reported side effects in the conventional group) and patient satisfaction (higher percentage of satisfied patients in the homeopathic group).

Conclusion

Overall patient satisfaction was significantly higher in homeopathic than in conventional care. Homeopathic treatments were perceived as a low-risk therapy with two to three times fewer side effects than conventional care

Patient satisfaction of primary care for musculoskeletal diseases: a comparison between Neural Therapy and conventional medicine

Background

The main objective of this study was to assess and compare patient satisfaction with Neural Therapy (NT) and conventional medicine (COM) in primary care for musculoskeletal diseases.

Methods

A cross-sectional study in primary care for musculoskeletal disorders covering 77 conventional primary care providers and 18 physicians certified in NT with 241 and 164 patients respectively. Patients and physicians documented consultations and patients completed questionnaires at a one-month follow-up. Physicians documented duration and severity of symptoms, diagnosis, and procedures. The main outcomes in the evaluation of patients were: fulfillment of expectations, perceived treatment effects, and patient satisfaction.

Results

The most frequent diagnoses belonged to the group of dorsopathies (39% in COM, 46% in NT). We found significant differences between NT and COM with regard to patient evaluations. NT patients documented better fulfilment of treatment expectations and higher overall treatment satisfaction. More patients in NT reported positive side effects and less frequent negative effects than patients in COM. Also, significant differences between NT and COM patients were seen in the quality of the patient-physician interaction (relation and communication, medical care, information and support, continuity and cooperation, facilities availability, and accessibility), where NT patients showed higher satisfaction. Differences were also found with regard to the physicians' management of disease, with fewer work incapacity attestations issued and longer consultation times in NT.

Conclusion

Our findings show a significantly higher treatment and care-related patient satisfaction with primary care for musculoskeletal diseases provided by physicians practising Neural Therapy.

Patient satisfaction with primary care: a comparison between conventional care and traditional Chinese medicine

OBJECTIVES

The evaluation of patient satisfaction with traditional Chinese medicine (TCM) versus conventional medicine (COM) in Swiss primary care.

DESIGN

A cross-sectional study was performed with questionnaires aimed at fulfilment of expectations, perceived treatment effects, and patient satisfaction.

SETTING

Participants were 51 certificated TCM physicians, 71 COM physicians, and 2530 adult patients.

RESULTS

Among patients seeking primary care in Switzerland, those who choose a conventional physician with additional certification in TCM have a higher chance of being completely satisfied with their treatment than patients who choose a physician educated only in COM.

CONCLUSIONS

Physicians who supply TCM in addition to COM are able to satisfy the needs of their patients more completely than solely COM-practicing physicians. Explanations for this difference include less frequent side effects in TCM, better TCM physician-patient interaction with longer duration of consultation, and different treatment expectations of patients seeking TCM.

Laterality and asymmetry of endometriotic lesions

OBJECTIVE

To identify possible sources of heterogeneities in the estimation of the proportion of bilateral cases and of left-sided cases of endometriotic lesions.

DESIGN

We included 20 studies that reported estimated proportions, and examined the effect of sample size and the anatomic location of lesions on the heterogeneity using a mixed-effect logit regression model.

SETTING

Academic.

PATIENT(S)

None.

INTERVENTION(S)

None.

MAIN OUTCOME MEASURE(S)

None.

RESULT(S)

The sample size of these studies ranged from 64 to 1,407, with a median of 227 and a total sample size of 7,236 cases. There is substantial heterogeneity in the estimated proportion of both bilateral and left-sided cases. The estimated proportion of bilateral cases is positively associated with the sample size of the study, whereas that of left-sided cases is negatively associated with the sample size, irrespective of the anatomic locations of endometriotic lesions.

CONCLUSION(S)

There is an identifiable source of heterogeneity in proportion estimates, with the sample size being an apparent source. Although the precise causes for the sample size dependency are unclear, it is possible that the invasive nature of endometriotic lesions may eventually render most cases bilateral. Moreover, there are both promoting as well as mitigating or negating factors that contributing to the asymmetry of endometriotic lesions.

Markov transition models for binary repeated measures with ignorable and nonignorable missing values

Motivated by problems encountered in studying treatments for drug dependence, where repeated binary outcomes arise from monitoring biomarkers for recent drug use, this article discusses a statistical strategy using Markov transition model for analyzing incomplete binary longitudinal data. When the mechanism giving rise to missing data can be assumed to be ;ignorable', standard Markov transition models can be applied to observed data to draw likelihood-based inference on transition probabilities between outcome events. Illustration of this approach is provided using binary results from urine drug screening in a clinical trial of baclofen for cocaine dependence. When longitudinal data have ;nonignorable' missingness mechanisms, random-effects Markov transition models can be used to model the joint distribution of the binary data matrix and the matrix of missingness indicators. Categorizing missingness patterns into those for occasional or ;intermittent' missingness and those for monotonic missingness or ;missingness due to dropout', the random-effects Markov transition model was applied to a data set containing repeated breath samples analyzed for expired carbon monoxide levels among opioid-dependent, methadone-maintained cigarette smokers in a smoking cessation trial. Markov transition models provide a novel reconceptualization of treatment outcomes, offering both intuitive statistical values and relevant clinical insights.

The prevalence of endometriosis in women with chronic pelvic pain

BACKGROUND

The 2004 American College of Obstetrics and Gynecology clinical management guideline states that the prevalence of endometriosis is approximately 33% in women with chronic pelvic pain (CPP). This estimate came from a review showing that 28% of adult women with CPP were found to have endometriosis. The prevalence of 28% in adult women was arrived based on a compilation of 11 published studies. Yet even within the 11 studies, the reported prevalence of endometriosis varies wildly, ranging from 2 to 74%. Such an astounding variation or heterogeneity raises the question whether it is appropriate to use a single prevalence of endometriosis for all women with CPP.

METHODS

We sought to identify possible sources of heterogeneities in the estimation of prevalence of endometriosis in women with CPP. We included more studies that reported prevalence estimates than the review, and examined the effect of sample size and the year of publication on the heterogeneity.

RESULTS

The year of publication is positively associated with the prevalence estimate, which may indicate an increasing awareness of various appearances of endometriosis, or the prevalence of endometriosis may have increased among women with CPP. An alternative analysis with removal of four studies reporting highest prevalence estimates indicated that sample size is negatively associated with the prevalence estimates while the year of publication became only marginally significant.

CONCLUSIONS

There are identifiable sources of heterogeneity in prevalence estimates, with the year of publication, sample size, and difference in evaluation of CPP being three apparent sources. Having a single prevalence estimate for all women with CPP may be too simplistic at best. The true prevalence is very likely to be higher than 33%.

Using SAS PROC NLMIXED to fit item response theory models

Researchers routinely construct tests or questionnaires containing a set of items that measure personality traits, cognitive abilities, political attitudes, and so forth. Typically, responses to these items are scored in discrete categories, such as points on a Likert scale or a choice out of several mutually exclusive alternatives. Item response theory (IRT) explains observed responses to items on a test (questionnaire) by a person's unobserved trait, ability, or attitude. Although applications of IRT modeling have increased considerably because of its utility in developing and assessing measuring instruments, IRT modeling has not been fully integrated into the curriculum of colleges and universities, mainly because existing general purpose statistical packages do not provide built-in routines with which to perform IRT modeling. Recent advances in statistical theory and the incorporation of those advances into general purpose statistical software such as the Statistical Analysis System (SAS) allow researchers to analyze measurement data by using a class of models known as generalized linear mixed effects models (McCulloch & Searle, 2001), which include IRT models as special cases. The purpose of this article is to demonstrate the generality and flexibility of using SAS to estimate IRT model parameters. With real data examples, we illustrate the implementations of a variety of IRT models for dichotomous, polytomous, and nominal responses. Since SAS is widely available in educational institutions, it is hoped that this article will contribute to the spread of IRT modeling in quantitative courses.

A Meta-Analysis of Primate Hand Preferences, Particularly for Reaching

P. F. MacNeilage, M. G. Studdert-Kennedy, and B. Lindblom (1987) proposed a progression for handedness in primates that was supposed to account for the evolution of a right bias in human handedness. To test this proposal, the authors performed meta-analyses on 62 studies that provided individual data (representing 31 species: 9 prosimians, 6 New World monkeys, 10 Old World monkeys, 2 lesser apes, and 4 greater apes), of the 118 studies of primate handedness published since 1987. Although evidence of a population-level left-handed bias for prosimians and Old World monkeys supports P. F. MacNeilage et al., the data from apes, New World monkeys, and individual species of prosimians and New World monkeys do not. Something other than primate handedness may have been the evolutionary precursor of the right bias in hand-use distribution among hominids.