On Graphically Checking Goodness-of-fit of Binary Logistic Regression Models

OBJECTIVES: This paper is concerned with checking goodness-of-fit of binary logistic regression models. For the practitioners of data analysis, the broad classes of procedures for checking goodness-of-fit available in the literature are described. The challenges of model checking in the context of binary logistic regression are reviewed. As a viable solution, a simple graphical procedure for checking goodness-of-fit is proposed. METHODS: The graphical procedure proposed relies on pieces of information available from any logistic analysis; the focus is on combining and presenting these in an informative way. RESULTS: The information gained using this approach is presented with three examples. In the discussion, the proposed method is put into context and compared with other graphical procedures for checking goodness-of-fit of binary logistic models available in the literature. CONCLUSION: A simple graphical method can significantly improve the understanding of any logistic regression analysis and help to prevent faulty conclusions.

LOGITCPRPLOT: Stata module to graph component-plus-residual plot for logistic regression

logitcprplot can be used after logistic regression for graphing a component-plus-residual plot (a.k.a. partial residual plot) for a given predictor, including a lowess, local polynomial, restricted cubic spline, fractional polynomial, penalized spline, regression spline, running line, or adaptive variable span running line smooth

RRLOGIT: Stata module to estimate logistic regression for randomized response data

rrlogit fits a maximum-likelihood logistic regression for randomized response data.

Predictors and overestimation of recalled mobile phone use among children and adolescents

A growing body of literature addresses possible health effects of mobile phone use in children and adolescents by relying on the study participants' retrospective reconstruction of mobile phone use. In this study, we used data from the international case-control study CEFALO to compare self-reported with objectively operator-recorded mobile phone use. The aim of the study was to assess predictors of level of mobile phone use as well as factors that are associated with overestimating own mobile phone use. For cumulative number and duration of calls as well as for time since first subscription we calculated the ratio of self-reported to operator-recorded mobile phone use. We used multiple linear regression models to assess possible predictors of the average number and duration of calls per day and logistic regression models to assess possible predictors of overestimation. The cumulative number and duration of calls as well as the time since first subscription of mobile phones were overestimated on average by the study participants. Likelihood to overestimate number and duration of calls was not significantly different for controls compared to cases (OR=1.1, 95%-CI: 0.5 to 2.5 and OR=1.9, 95%-CI: 0.85 to 4.3, respectively). However, likelihood to overestimate was associated with other health related factors such as age and sex. As a consequence, such factors act as confounders in studies relying solely on self-reported mobile phone use and have to be considered in the analysis.

Epidemiology and predictors of spinal injury in adult major trauma patients: European cohort study

This is a European cohort study on predictors of spinal injury in adult (≥16 years) major trauma patients, using prospectively collected data of the Trauma Audit and Research Network from 1988 to 2009. Predictors for spinal fractures/dislocations or spinal cord injury were determined using univariate and multivariate logistic regression analysis. 250,584 patients were analysed. 24,000 patients (9.6%) sustained spinal fractures/dislocations alone and 4,489 (1.8%) sustained spinal cord injury with or without fractures/dislocations. Spinal injury patients had a median age of 44.5 years (IQR = 28.8-64.0) and Injury Severity Score of 9 (IQR = 4-17). 64.9% were male. 45% of patients suffered associated injuries to other body regions. Age <45 years (≥45 years OR 0.83-0.94), Glasgow Coma Score (GCS) 3-8 (OR 1.10, 95% CI 1.02-1.19), falls >2 m (OR 4.17, 95% CI 3.98-4.37), sports injuries (OR 2.79, 95% CI 2.41-3.23) and road traffic collisions (RTCs) (OR 1.91, 95% CI 1.83-2.00) were predictors for spinal fractures/dislocations. Age <45 years (≥45 years OR 0.78-0.90), male gender (female OR 0.78, 95% CI 0.72-0.85), GCS <15 (OR 1.36-1.93), associated chest injury (OR 1.10, 95% CI 1.01-1.20), sports injuries (OR 3.98, 95% CI 3.04-5.21), falls >2 m (OR 3.60, 95% CI 3.21-4.04), RTCs (OR 2.20, 95% CI 1.96-2.46) and shooting (OR 1.91, 95% CI 1.21-3.00) were predictors for spinal cord injury. Multilevel injury was found in 10.4% of fractures/dislocations and in 1.3% of cord injury patients. As spinal trauma occurred in >10% of major trauma patients, aggressive evaluation of the spine is warranted, especially, in males, patients <45 years, with a GCS <15, concomitant chest injury and/or dangerous injury mechanisms (falls >2 m, sports injuries, RTCs and shooting). Diagnostic imaging of the whole spine and a diligent search for associated injuries are substantial.

Frequency of and predictors for patient-reported medical and medication errors in Switzerland

OBJECTIVES: To analyse the frequency of and identify risk factors for patient-reported medical errors in Switzerland. The joint effect of risk factors on error-reporting probability was modelled for hypothetical patients. METHODS: A representative population sample of Swiss citizens (n = 1306) was surveyed as part of the Commonwealth Fund’s 2010 lnternational Survey of the General Public’s Views of their Health Care System’s Performance in Eleven Countries. Data on personal background, utilisation of health care, coordination of care problems and reported errors were assessed. Logistic regression analysis was conducted to identify risk factors for patients’ reports of medical mistakes and medication errors. RESULTS: 11.4% of participants reported at least one error in their care in the previous two years (8% medical errors, 5.3% medication errors). Poor coordination of care experiences was frequent. 7.8% experienced that test results or medical records were not available, 17.2% received conflicting information from care providers and 11.5% reported that tests were ordered although they had been done before. Age (OR = 0.98, p = 0.014), poor health (OR = 2.95, p = 0.007), utilisation of emergency care (OR = 2.45, p = 0.003), inpatient-stay (OR = 2.31, p = 0.010) and poor care coordination (OR = 5.43, p <0.001) are important predictors for reporting error. For high utilisers of care that unify multiple risk factors the probability that errors are reported rises up to p = 0.8. CONCLUSIONS: Patient safety remains a major challenge for the Swiss health care system. Despite the health related and economic burden associated with it, the widespread experience of medical error in some subpopulations also has the potential to erode trust in the health care system as a whole.

Predictors of functional decline in elderly patients undergoing transcatheter aortic valve implantation (TAVI)

Aims This study aimed to assess functional course in elderly patients undergoing transcatheter aortic valve implantation (TAVI) and to find predictors of functional decline. Methods and results In this prospective cohort, functional course was assessed in patients ≥70 years using basic activities of daily living (BADL) before and 6 months after TAVI. Baseline EuroSCORE, STS score, and a frailty index (based on assessment of cognition, mobility, nutrition, instrumental and basic activities of daily living) were evaluated to predict functional decline (deterioration in BADL) using logistic regression models. Functional decline was observed in 22 (20.8%) of 106 surviving patients. EuroSCORE (OR per 10% increase 1.18, 95% CI: 0.83-1.68, P = 0.35) and STS score (OR per 5% increase 1.64, 95% CI: 0.87-3.09, P = 0.13) weakly predicted functional decline. In contrast, the frailty index strongly predicted functional decline in univariable (OR per 1 point increase 1.57, 95% CI: 1.20-2.05, P = 0.001) and bivariable analyses (OR: 1.56, 95% CI: 1.20-2.04, P = 0.001 controlled for EuroSCORE; OR: 1.53, 95% CI: 1.17-2.02, P = 0.002 controlled for STS score). Overall predictive performance was best for the frailty index [Nagelkerke's R(2) (NR(2)) 0.135] and low for the EuroSCORE (NR(2) 0.015) and STS score (NR(2) 0.034). In univariable analyses, all components of the frailty index contributed to the prediction of functional decline. Conclusion Over a 6-month period, functional status worsened only in a minority of patients surviving TAVI. The frailty index, but not established risk scores, was predictive of functional decline. Refinement of this index might help to identify patients who potentially benefit from additional geriatric interventions after TAVI.

Predictors of sickness absence in patients with a new episode of low back pain in primary care

This study examines predictors of sickness absence in patients presenting to a health practitioner with acute/ subacute low back pain (LBP). Aims of this study were to identify baseline-variables that detect patients with a new LBP episode at risk of sickness absence and to identify prognostic models for sickness absence at different time points after initial presentation. Prospective cohort study investigating 310 patients presenting to a health practitioner with a new episode of LBP at baseline, three-, six-, twelve-week and six-month follow-up, addressing work-related, psychological and biomedical factors. Multivariate logistic regression analysis was performed to identify baseline-predictors of sickness absence at different time points. Prognostic models comprised 'job control', 'depression' and 'functional limitation' as predictive baseline-factors of sickness absence at three and six-week follow-up with 'job control' being the best single predictor (OR 0.47; 95%CI 0.26-0.87). The six-week model explained 47% of variance of sickness absence at six-week follow-up (p<0.001). The prediction of sickness absence beyond six-weeks is limited, and health practitioners should re-assess patients at six weeks, especially if they have previously been identified as at risk of sickness absence. This would allow timely intervention with measures designed to reduce the likelihood of prolonged sickness absence.

Predictors of substance use reduction in an epidemiological first-episode psychosis cohort

Aim To assess the predictors of a significant decrease or cessation of substance use (SU) in a treated epidemiological cohort of first-episode psychosis (FEP) patients. Method Participants were FEP patients of the Early Psychosis Prevention and Intervention Centre in Australia. Patients' medical files were reviewed using a standardized file audit. Data on 432 patients with FEP and baseline co-morbid substance use disorder (SUD) were available for analysis. Predictors of reduction/cessation of SU at follow up were examined using logistic regression analyses. Results In univariate analyses, a reduction/cessation of SU was predicted by baseline measures reflecting higher education, employment, accommodation with others, cannabis use disorder (CUD) only (rather than poly-SUDs), better global functioning and better premorbid social and occupational functioning, later age at onset of psychosis, and a diagnosis of non-affective psychosis. In multivariate analysis, CUD alone and better premorbid social and occupational functioning remained significant predictors. Conclusions Addressing SUDs and social and occupational goals in people with FEP may offer opportunities to prevent SUDs becoming more severe or entrenched. Further longitudinal research on recovery from SU and FEP is needed to disentangle directions of influence and identify key targets for intervention.

Are there accurate predictors of long-term vital and functional outcomes in cardiac surgical patients requiring prolonged intensive care?

BACKGROUND AND OBJECTIVE: The decision to maintain intensive treatment in cardiac surgical patients with poor initial outcome is mostly based on individual experience. The risk scoring systems used in cardiac surgery have no prognostic value for individuals. This study aims to assess (a) factors possibly related to poor survival and functional outcomes in cardiac surgery patients requiring prolonged (> or = 5 days) intensive care unit (ICU) treatment, (b) conditions in which treatment withdrawal might be justified, and (c) the patient's perception of the benefits and drawbacks of long intensive treatments. METHODS: The computerized data prospectively recorded for every patient in the intensive care unit over a 3-year period were reviewed and analyzed (n=1859). Survival and quality of life (QOL) outcomes were determined in all patients having required > or =5 consecutive days of intensive treatment (n=194/10.4%). Long-term survivors were interviewed at yearly intervals in a standardized manner and quality of life was assessed using the dependency score of Karnofsky. No interventions or treatments were given, withhold, or withdrawn as part of this study. RESULTS: In-hospital, 1-, and 3-year cumulative survival rates reached 91.3%, 85.6%, and 75.1%, respectively. Quality of life assessed 1 year postoperatively by the score of Karnofsky was good in 119/165 patients, fair in 32 and poor in 14. Multivariate logistic regression analysis of 19 potential predictors of poor outcome identified dialysis as the sole factor significantly (p=0.027) - albeit moderately - reducing long-term survival, and sustained neurological deficit as an inconstant predictor of poor functional outcome (p=0.028). One year postoperatively 0.63% of patients still reminded of severe suffering in the intensive station and 20% of discomfort. Only 7.7% of patients would definitely refuse redo surgery. CONCLUSIONS: This study of cardiac surgical patients requiring > or =5 days of intensive treatment did not identify factors unequivocally justifying early treatment limitation in individuals. It found that 1-year mortality and disability rates can be maintained at a low level in this subset of patients, and that severe suffering in the ICU is infrequent.

Clinical and radiographic assessment of various predictors for healing outcome 1 year after periapical surgery

This clinical study prospectively evaluated the influence of various predictors on healing outcome 1 year after periapical surgery. The study cohort included 194 teeth in an equal number of patients. Three teeth were lost for the follow-up (1.5% drop-out rate). Clinical and radiographic measures were used to determine the healing outcome. For statistical analysis, results were dichotomized (healed versus nonhealed). The overall success rate was 83.8% (healed cases). The only individual predictors to prove significant for the outcome were pain at initial examination (p=0.030) and other clinical signs or symptoms at initial examination (p=0.042), meaning that such teeth had lower healing rates 1 year after periapical surgery compared with teeth without such signs or symptoms. Logistic regression revealed that pain at initial examination (odds ratio=2.59, confidence interval=1.2-5.6, p=0.04) was the only predictor reaching significance. Several predictors almost reached statistical significance: lesion size (p=0.06), retrofilling material (p=0.06), and postoperative healing course (p=0.06).

Implant Failure Predictors in the Posterior Maxilla: A Retrospective Study of 273 Consecutive Implants

Background: The goal of this study was to retrospectively analyze a cohort of 136 patients who underwent dental implant placement in the posterior maxilla at the University of Connecticut Health Center to assess and identify predictors for implant failure in the posterior maxilla. Methods: Data were retrieved from patient charts to identify subjects older than 21 years of age who received dental implant(s) in the posterior maxilla. Patients without a postoperative baseline radiograph were excluded. A recall radiograph was taken 3 to 6 months after implant placement. If there was no recall radiograph, the subject was contacted for a recall visit that included a clinical evaluation and radiographs to determine the implant status. Based on a univariate screening, variables considered potential implant failure predictors included gender, diabetes, smoking, implant length, implant diameter, membrane use, sinus-elevation technique, and surgical complications. These parameters were further assessed, and a multivariable logistic regression was performed with implant failure as a dependant variable. All tests of significance were evaluated at the 0.05 error level. Results: Two hundred seventy-three implants were placed in the posterior maxilla. Fourteen implants failed (early and late failures combined), resulting in a 94.9% overall survival rate. The survival rates for the sinus-elevation group and native bone group were 92.2% and 96.7%, respectively (P = 0.090). Based on the multivariable analysis, sinus floor-elevation procedures were not associated with increased risk for implant failure (P = 0.702). In contrast, smoking and surgical complications had a statistically significant effect on implant failure; the odds ratios for implant failure were 6.4 (P = 0.025) and 8.2 (P = 0.004), respectively. Conclusion: Sinus-elevation procedures with simultaneous or staged implant placement do not increase the risk for implant failure, whereas smoking and surgical complications markedly increase the risk for implant failure.

Predictors of Weight Loss Maintenance following an Insurance-Sponsored Weight Management Program

Intentional weight loss among overweight and obese adults (body mass index ≥ 25 kg/m2) is associated with numerous health benefits, but weight loss maintenance (WLM) following participation in weight management programming has proven to be elusive. Many individuals attempting to lose weight join formal programs, especially women, but these programs vary widely in focus, as do postprogram weight regain results. We surveyed 2,106 former participants in a community-based, insurance-sponsored weight management program in the United States to identify the pre, during, and post-intervention behavioral and psychosocial factors that lead to successful WLM. Of 835 survey respondents (39.6% response rate), 450 met criteria for inclusion in this study. Logistic regression analyses suggest that interventionists should assess and discuss weight loss and behavior change perceptions early in a program. However, in developing maintenance plans later in a program, attention should shift to behaviors, such as weekly weighing, limiting snacking in the evening, limiting portion sizes, and being physically active every day.

Kidney disease in antiretroviral-naïve HIV-positive adults with high CD4 counts: prevalence and predictors of kidney disease at enrolment in the INSIGHT Strategic Timing of AntiRetroviral Treatment (START) trial.

OBJECTIVES HIV infection has been associated with an increased risk of chronic kidney disease (CKD). Little is known about the prevalence of CKD in individuals with high CD4 cell counts prior to initiation of antiretroviral therapy (ART). We sought to address this knowledge gap. METHODS We describe the prevalence of CKD among 4637 ART-naïve adults (mean age 36.8 years) with CD4 cell counts > 500 cells/μL at enrolment in the Strategic Timing of AntiRetroviral Treatment (START) study. CKD was defined by estimated glomerular filtration rate (eGFR) < 60 mL/min/1.73 m(2) and/or dipstick urine protein ≥ 1+. Logistic regression was used to identify baseline characteristics associated with CKD. RESULTS Among 286 [6.2%; 95% confidence interval (CI) 5.5%, 6.9%] participants with CKD, the majority had isolated proteinuria. A total of 268 participants had urine protein ≥ 1+, including 41 with urine protein ≥ 2+. Only 22 participants (0.5%) had an estimated glomerular filtration rate < 60 mL/min/1.73 m(2) , including four who also had proteinuria. Baseline characteristics independently associated with CKD included diabetes [adjusted odds ratio (aOR) 1.73; 95% CI 1.05, 2.85], hypertension (aOR 1.82; 95% CI 1.38, 2.38), and race/ethnicity (aOR 0.59; 95% CI 0.37, 0.93 for Hispanic vs. white). CONCLUSIONS We observed a low prevalence of CKD associated with traditional CKD risk factors among ART-naïve clinical trial participants with CD4 cell counts > 500 cells/μL.