Immortal time bias in observational studies of time-to-event outcomes

PurposeThe purpose of the study is to show, through simulation and example, the magnitude and direction of immortal time bias when an inappropriate analysis is used.Materials and methodsWe compare 4 methods of analysis for observational studies of time-to-event outcomes: logistic regression, standard Cox model, landmark analysis, and time-dependent Cox model using an example data set of patients critically ill with influenza and a simulation study.ResultsFor the example data set, logistic regression, standard Cox model, and landmark analysis all showed some evidence that treatment with oseltamivir provides protection from mortality in patients critically ill with influenza. However, when the time-dependent nature of treatment exposure is taken account of using a time-dependent Cox model, there is no longer evidence of a protective effect of treatment. The simulation study showed that, under various scenarios, the time-dependent Cox model consistently provides unbiased treatment effect estimates, whereas standard Cox model leads to bias in favor of treatment. Logistic regression and landmark analysis may also lead to bias.ConclusionsTo minimize the risk of immortal time bias in observational studies of survival outcomes, we strongly suggest time-dependent exposures be included as time-dependent variables in hazard-based analyses.     

Time-Dependent Impact of Diabetes on Mortality in Patients With Stroke: Survival up to 5 years in a health insurance population cohort in Germany

OBJECTIVE To estimate the impact of diabetes on mortality in patients after first stroke event. RESEARCH DESIGN AND METHODS Using claims data from a nationwide statutory health insurance fund (Gmünder ErsatzKasse), we assessed all deaths in a cohort of 5,757 patients with a first stroke between 2005 and 2007 (69.3% male, mean age 68.1 years, 32.2% with diabetes) up to 2009. By use of Cox regression, we estimated time-dependent hazard ratios (HRs) to compare patients with and without diabetes stratified by sex. RESULTS The cumulative 5-year mortality was 40.0 and 54.2% in diabetic men and women, and 32.3 and 38.1% in their nondiabetic counterparts, respectively. In males, mortality was significantly lower in diabetic compared with nondiabetic patients in the first 30 days (multiple-adjusted HR 0.67 [95% CI 0.53-0.84]). After approximately a quarter of a year, the diabetes risk increased, yielding crossed survival curves. Later on, mortality risk tended to be similar in diabetic and nondiabetic men (1-2 years: 1.42 [1.09-1.85]; 3-5 years: 1.00 [0.67-1.41]; time dependency of diabetes, P = 0.008). In women, the pattern was similar; however, time dependency was not statistically significant (P = 0.89). Increasing age, hemorrhagic stroke, renal failure (only in men), levels of care dependency, and number of prescribed medications were significantly associated with mortality. CONCLUSIONS We found a time-dependent mortality risk of diabetes after first stroke in men. Possible explanations may be type of stroke or earlier and more intensive treatment of risk factors in diabetic patients.     

Predictive survival model with time-dependent prognostic factors: development of computer-aided SAS Macro program

AIMS AND OBJECTIVES: Computer program for the prediction of survival with respect to time-dependent proportional hazards regression model has been rarely addressed. We therefore developed a SAS Macro program for time-dependent Cox regression predictive model for empirical survival data associated with time-dependent covariates. METHOD: Time-dependent proportional hazards regression model and partial likelihood in association with time-varying predictors were explicitly delineated. Baseline hazard using Andersen's method was incorporated into proportional hazards regression model to predict the dynamic change of cumulative survival in respect of time-varying predictors. Two SAS Macro programs for time-dependent predictive survival model and model validation using receiver operative characteristics were written with SAS IML language. RESULTS: The computer program was applied to data on clinical surveillance of small hepatocellular carcinoma (HCC) treated by percutaneous ethanol injection (PEI) or transcatheter arterial embolization (TAE) with time-varying predictors such as alpha-feto protein (AFP) and other biological markers. CONCLUSION: The program is very useful for real-time prediction of cumulative survival on the basis of time-dependent covariates.     

Prompt antibiotic administration and goal-directed hemodynamic support in patients with severe sepsis and septic shock

PURPOSE OF REVIEW: Treatment protocols targeting the rapid administration of appropriate antibiotics and hemodynamic support are now recognized as a key measure in the initial care of patients presenting with severe sepsis and septic shock. Strong evidence exists showing that time parameters, particularly in the emergency department, are as important as the nature of the treatment administered. The concept of sepsis bundles integrates evidence-based and time-sensitive issues, derived from international sepsis guidelines, to ensure that all eligible patients receive the right treatment as early as possible. RECENT FINDINGS: Several studies have demonstrated that patients resuscitated according to sepsis bundles had a significantly lower mortality. SUMMARY: It seems logical that timely and protocolized treatment for patients presenting with severe sepsis and septic shock will impact on outcome. It remains to be shown, however, whether translating evidence into clinical practice will increase adherence to the bundles and positively impact on survival.     

Unintended bias, clinical trial results, and the heparin post hoc crossover fallacy

OBJECTIVE: To describe the unintended pitfalls in the interpretation of postrandomization events in clinical trials. DESIGN: Analysis of patients enrolled in clinical trials for new sepsis interventions with postrandomization exposure to heparin. PATIENTS: Retrospective review of patients enrolled in large phase III sepsis trials after treatment with experimental anticoagulant therapies. INTERVENTIONS: Nonrandomized exposure to heparin therapy administered for a variety of clinical indications after enrollment in large phase III sepsis trials. RESULTS: The effect of heparin on overall survival in septic patients in trials that randomized patients into treatment assignments other than heparin is difficult to quantitatively analyze because of unintended selection bias and allocation bias. Both forms of bias overestimate the potential therapeutic value of heparin. This post hoc analysis of the data is functionally a crossover study in which only surviving patients can cross over in one direction (toward the heparin treatment arm). CONCLUSION: Great caution should be exercised in the post hoc interpretation of the potential efficacy of nonrandomized treatments such as heparin therapy derived from phase III clinical data of other drugs for sepsis. The therapeutic value of heparin as a treatment modality in severe sepsis can best be determined in a formal, randomized, prospective clinical trial. This will obviate the unavoidable selection bias and allocation bias intrinsic to postrandomization events in clinical trials with a high early mortality rate such as severe sepsis and septic shock.     

Reducing racial disparities in pain treatment: the role of empathy and perspective-taking

Epidemiological evidence indicates that African Americans receive lower quality pain treatment than European Americans. However, the factors causing these disparities remain unidentified, and solutions to this problem remain elusive. Across three laboratory experiments, we examined the hypotheses that empathy is not only causing pain treatment disparities but that empathy-inducing interventions can reduce these disparities. Undergraduates (Experiments 1 and 2) and nursing professionals (Experiment 3) watched videos of real Black and White patients’ genuine facial expressions of pain, provided pain treatment decisions, and reported their feelings of empathy for each patient. The efficacy of an empathy-inducing, perspective-taking intervention at reducing pain treatment disparities was also examined (Experiments 2 and 3). When instructed to attempt to provide patients with the best care, participants exhibited significant pro-White pain treatment biases. However, participants engaged in an empathy-inducing, perspective-taking intervention that instructed them to imagine how patients’ pain affected patients’ lives exhibited upwards of a 55% reduction in pain treatment bias in comparison to controls. Furthermore, Pro-White empathy biases were highly predictive of pro-White pain treatment biases. The magnitude of the empathy bias experienced predicted the magnitude of the treatment bias exhibited. These findings suggest that empathy plays a crucial role in racial pain treatment disparities in that it appears not only to be one likely cause of pain treatment disparities but also is an important means for reducing racial disparities in pain treatment.     

Is there a potential role for attention bias modification in pain patients? Results of 2 randomised,controlled trials

Potential applications of attention bias modification (ABM) for acute and chronic pain patients are investigated. In study 1, 54 acute back pain patients (46 of whom completed the study) were recruited at their initial physiotherapy session and randomised to receive 1 session of ABM or placebo. Patients were followed up 3 months later. Participants who were randomised to receive ABM reported less average (P=0.001) and current pain (P=0.008) and experienced pain for fewer days (P=0.01) than those who received placebo. In study 2, 34 chronic pain patients were recruited and randomly assigned to receive either 4 sessions of ABM (n=22) or placebo (n=12), followed by 8 sessions of cognitive behavioural treatment (CBT). After ABM, there was a significant group-by-time effect for disability. By 6-month follow-up, differences had emerged between the 2 training groups, such that the ABM group had shown greater reductions in anxiety sensitivity and disability than the placebo group. Although the results of these studies show that there is potential in the application of ABM to pain conditions, the mechanisms of treatment could not be established. Neither group showed an initial bias towards the word stimuli or a training effect, and only in the acute pain group were changes in biases related to outcome. Nonetheless, the fact that 2 independent samples showed a positive effect of ABM on clinical outcomes suggests that ABM is worthy of future study as an intervention for pain patients.     

Outcome and prognostic features of intensive care unit treatment in patients with hematological malignancies

OBJECTIVE: To assess the outcome of intensive care unit (ICU) treatment in patients with hematological malignancies. DESIGN AND SETTING: Retrospective cohort study in the medical ICU of a university hospital. PATIENTS: 104 critically ill patients after receiving conventional chemotherapy or autologous hematopoietic stem cell transplantation. INTERVENTIONS: We analyzed demographic data, underlying disease, intensity of antineoplastic regimen, cause of admission, need for mechanical ventilation, and hemofiltration, ICU survival, and survival after discharge, furthermore neutrophil count, C-reactive protein (150 mg/l), antithrombin III, prothrombin time, and SAPS II (50) at ICU admission. All recorded variables were evaluated for prognostic relevance by univariate and multivariate analyses. MEASUREMENTS AND RESULTS: Overall ICU mortality was 44%, with significantly higher mortality in ventilated patients (74% vs. 12% in nonventilated patients, p<0.001). Overall survival for the entire group 6 months and 1 year after ICU admission was 33% and 29%, respectively. Multivariate analysis revealed mechanical ventilation and SAPS II as independent prognostic factors of both ICU mortality and long-term survival, while C-reactive protein predicted only ICU mortality. CONCLUSIONS: The outcome of patients not requiring ventilatory support in this study was encouraging, while invasive ventilation was again confirmed as predicting a dismal prognosis in this population. Efforts should be directed to avoiding this procedure by reducing the pulmonary toxicity of antineoplastic treatment and to making ventilatory support more tolerable.     

High-dose antithrombin III in the treatment of severe sepsis in patients with a high risk of death: efficacy and safety

OBJECTIVE: To explore if patients with severe sepsis and with a predicted high risk of death (according to the Simplified Acute Physiology Score II) might have a treatment benefit from high-dose antithrombin III. DESIGN: Subgroup analysis of a randomized, placebo-controlled, double-blind, prospective phase III study. SETTING: Unifactorial and multifactorial reanalysis of prospectively defined populations from the KyberSept trial. PATIENTS: We studied 1,008 patients (43.6% of the overall intention-to-treat population, n = 2,314) with a predicted mortality rate of 30-60% at study entry as defined by the Simplified Acute Physiology Score II. INTERVENTIONS: Patients were randomized in a 1:1 fashion to receive either high-dose antithrombin III (30,000 IU intravenously over the period of 4 days) or placebo. MEASUREMENTS AND MAIN RESULTS: In a Kaplan-Meier analysis of patients with a predicted mortality of 30-60%, the survival time when followed up for 90 days after admission was increased in the high-dose antithrombin III group compared with placebo (p = .04). If heparin was avoided during the 4-day treatment phase with high-dose antithrombin III (n = 140) or placebo (n = 162), the treatment effect appeared to be even more pronounced: 28-day mortality rate, 35.7% vs. 44.4% (risk ratio, 0.804; 95% confidence interval, 0.607-1.064); 56-day mortality rate, 39.9% vs. 52.2% (risk ratio, 0.764; 95% confidence interval, 0.593-0.984); 90-day mortality rate, 42.8% vs. 55.1% (risk ratio, 0.776; 95% confidence interval, 0.614-0.986). Like in the overall population, the percentage with any bleeding was increased in patients receiving high-dose antithrombin III compared with placebo. Survival rates were in favor of high-dose antithrombin III in patients both with and without bleeding complications. CONCLUSIONS: Treatment with high-dose antithrombin III may increase survival time up to 90 days in patients with severe sepsis and high risk of death. This benefit may even be stronger when concomitant heparin is avoided.     

Monitoring of intracranial pressure in patients with severe traumatic brain injury: an Austrian prospective multicenter study

OBJECTIVE: The goals of this study were to elucidate reasons why patients did or did not receive intracranial pressure (ICP) monitoring and to describe factors influencing hospital mortality after severe traumatic brain injury (TBI). DESIGN: Prospective multicenter cohort study. PATIENTS AND PARTICIPANTS: 88,274 patients consecutively admitted to 32 medical, surgical and mixed Austrian ICUs between 1998 and 2004. INTERVENTIONS: None. MEASUREMENTS AND RESULTS: 1,856 patients (2.1% of all ICU admissions) exhibited severe TBI (GCS < 9); of these, 1,031 (56%) had ICP monitoring. The "worst" and the "best" cases were both less likely to receive ICP monitoring. Younger patients, female patients, and patients with isolated TBI were more likely to receive ICP monitoring. Compared with large centers ICP was monitored more frequently [odds ratio (OR) 3.09, CI 2.42-3.94] in medium-sized centers. The 20% of patients with the highest likelihood to receive ICP monitoring were monitored in 91% of cases, and had the lowest hospital mortality (31%, OR 0.78, CI 0.37-1.64). Multivariate analysis revealed that severity of illness, TBI severity, isolated TBI, and the number of cases treated per year were associated with hospital outcome. Compared with the large centers, ORs for hospital mortality were 1.85 (CI 1.42-2.40) for patients from medium-sized centers and 1.91 (CI 1.24-2.93) for patients from small centers. CONCLUSIONS: ICP monitoring may possibly have some beneficial effects, but this needs further evaluation. Patients with severe TBI should be admitted to experienced centers with high patient volumes since this might improve hospital mortality rates.     

Ego bias, reverse ego bias, and physicians' prognostic

OBJECTIVE: To evaluate the effects of "ego bias" on physicians' prognostic judgments. Ego bias is defined as systematic overestimation of the prognosis of one's own patients compared with the expected outcome of a population of similar patients. DESIGN: A prospective study of an inception cohort of critically ill patients followed until death or discharge from the hospital. PATIENTS: Consecutive patients admitted to either an ICU or an intermediate ICU at a teaching hospital during January and February 1987, excluding patients admitted after coronary artery bypass grafting, for elective dialysis, or transferred to the intermediate ICU from another critical care unit. MAIN OUTCOME MEASURES AND COMPARISONS: House officers' and critical care attending physicians' assessments of the likelihood of inhospital survival for each patient, and their assessments of the overall survival rate of ICU and intermediate ICU patients were compared with each other and with actual survival rates. RESULTS: The attending physicians' predictions for individual patients were significantly lower than their judgments of the overall survival rate, 79.8% vs. 88.0%, p = .0067, suggesting the presence of a "reverse ego bias." The house officers' predictions for individual patients were significantly higher than their judgments of the overall survival rate, 73.5% vs. 68.9%, p = .018, suggesting the presence of ego bias. The magnitude and directions of these differences varied significantly among the attending physicians (F = 4.3, degrees of freedom = 3, p = .0062 by repeated-measures analysis of variance) and the house officers (F = 6.3, degrees of freedom = 5, p = .0001). CONCLUSIONS: The critical care attending physicians exhibited reverse ego bias that was mainly a function of their optimism about the overall survival rate for critically ill patients. The house officers exhibited ego bias that was mainly a function of their pessimism about the overall survival rate for critically ill patients.     

Impact of diabetes on long-term survival after acute myocardial infarction: comparability of risk with prior myocardial infarction

OBJECTIVE: To determine the effect of diabetes on long-term survival after acute myocardial infarction and to compare its effect with that of a previous myocardial infarction. RESEARCH DESIGN AND METHODS: In a prospective cohort study, we followed 1,935 patients hospitalized with a confirmed acute myocardial infarction at 45 U.S. medical centers between 1989 and 1993, as part of the Determinants of Myocardial Infarction Onset Study. Trained interviewers performed chart reviews and face-to-face interviews with all patients. We analyzed survival using Cox proportional hazards regression to control for potentially confounding factors. RESULTS: Of the 1,935 patients, 320 (17%) died during a mean follow-up of 3.7 years. A total of 399 patients (21%) had previously diagnosed diabetes. Diabetes was associated with markedly higher total mortality in unadjusted (hazard ratio [HR] 2.4; 95% CI 1.9-3.0) and adjusted (1.7; 1.3-2.1) analyses. The magnitude of the effect of diabetes was identical to that of a previous myocardial infarction. The effect of diabetes was not significantly modified by age, smoking, household income, use of thrombolytic therapy, type of hypoglycemic treatment, or duration of diabetes, but the risk associated with diabetes was higher among women than men (adjusted HRs 2.7 vs. 1.3, P = 0.01). CONCLUSIONS: Diabetes is associated with markedly increased mortality after acute myocardial infarction, particularly in women. The increase in risk is of the same magnitude as a previous myocardial infarction and provides further support for aggressive treatment of coronary risk factors among diabetic patients.     

Re-analysis of survival data of cancer patients utilizing additive homeopathy

In this short communication we present a re-analysis of homeopathic patient data in comparison to control patient data from the same Outpatient´s Unit “Homeopathy in malignant diseases” of the Medical University of Vienna. In this analysis we took account of a probable immortal time bias. For patients suffering from advanced stages of cancer and surviving the first 6 or 12 months after diagnosis, respectively, the results show that utilizing homeopathy gives a statistically significant (p<0.001) advantage over control patients regarding survival time. In conclusion, bearing in mind all limitations, the results of this retrospective study suggest that patients with advanced stages of cancer might benefit from additional homeopathic treatment until a survival time of up to 12 months after diagnosis.     

Plasmapheresis in severe sepsis and septic shock: a prospective,randomised, controlled trial

OBJECTIVE: To determine the therapeutic efficacy and safety of plasmapheresis in the treatment of patients with severe sepsis and septic shock. DESIGN: Prospective, randomised, clinical trial with a planned, midstudy, interim analysis. SETTING: Intensive care unit in a university hospital in Archangels, Russia. PATIENTS: Consecutive patients with severe sepsis or septic shock. INTERVENTIONS: One hundred and six patients were randomised to receive either standard therapy or an add-on treatment with plasmapheresis. MEASUREMENTS AND RESULTS: The primary endpoint was 28-day survival. Septic shock was diagnosed in 57% of the plasmapheresis-treated patients and 54% of the control patients. Mean APACHE III score at entry was 56.4 in the plasmapheresis group and 53.5 in the control group. The 28-day, all-cause mortality rate was 33.3% (18/54) in the plasmapheresis group and 53.8% (28/52) in the control group. This represents a relative risk for fatal outcome in the plasmapheresis group of 0.61, an absolute risk reduction of 20.5% and a number of patients needed to treat of 4.9. Apart from six transient episodes of hypotension and one allergic reaction to fresh frozen plasma, no adverse reactions were attributable to the plasmapheresis treatment in this study. CONCLUSIONS: Plasmapheresis may be an important adjuvant to conventional treatment to reduce mortality in patients with severe sepsis or septic shock. Plasmapheresis is a safe procedure in the treatment of septic patients. A prospective randomised multicentre trial is warranted to confirm our results and to determine which subgroups of septic patients will benefit most from this treatment modality.     

Impact of neutropenia duration on short-term mortality in neutropenic critically ill cancer patients

OBJECTIVE: To identify predictors of 30-day mortality and to assess the impact of neutropenia recovery (NR) on 30-day mortality in critically ill cancer patients (CICPs). DESIGN AND SETTING: Retrospective review of the medical records of the 102 neutropenic CICPs admitted to a medical intensive care unit (ICU) over a 10-year period. INTERVENTION: None. MEASUREMENTS AND RESULTS: Malignancies consisted of acute leukemia (n=42), lymphoma (n=23), myeloma (n=28), and solid tumors (n=9). Reasons for ICU admission were acute respiratory failure (n=81), shock (n=58), acute renal failure (n=33), and coma (n=13). Seventy patients needed conventional mechanical ventilation (MV) and 21 noninvasive MV, 67 vasopressor agents, and 28 dialysis. Sixty-two patients experienced NR during their ICU stay. In a multivariate logistic regression model, 30-day mortality was higher in patients with acute respiratory or renal failure and lower in patients with NR (OR, 0.09 [0.01-0.86]). This model assumed that patients who experienced NR in the ICU were merely these who did not die early in the ICU. To take into account the effect of time to occurrence of NR on time to death we secondarily used a Cox model including neutropenia duration and NR as time-dependent variables. In this second model, the only significant predictors of 30-day mortality were age, respiratory failure, renal failure, and coma. CONCLUSION: Organ failure but not disease progression or neutropenia duration affect 30-day mortality in neutropenic CICPs. ICU-acquired events might be modeled as time-dependent variables in a Cox model, rather than standard covariates in logistic regression models.     

Palliative Care Clinician Overestimation of Survival in Advanced Cancer: Disparities and Association With End-of-Life Care

Context

Clinicians frequently overestimate survival time in serious illness.

The impact of record-linkage bias in the Cox model

Rationale, aims and objectives  Record linkage (RL) has become increasingly useful in health care administration, demographic studies, provision of health statistics and medical research. Linkage failure may occur when databases are affected by missing or inaccurate information. In particular, if the subsets of those who are not linked are not representative of the original population, the results obtained from linked data may be biased. This paper discusses the impact of incomplete RL on survival analysis.
Methods  In our study we assess by simulations the potential impact of such bias, that we will refer to as RL, on the effect of the covariates in the Cox regression model. We also evaluate the RL bias introduced by an incomplete linkage procedure on the analysis of survival in a cohort of patients with breast cancer.
Results  Our simulation study shows that the relative bias of the factors, which the linking probability depends on, reaches the threshold of 20%, and is never less than 5%. The bias observed in the simulation for a comparable scenario is consistent with the actual one estimated from the breast cancer records.
Conclusions  Incomplete RL is rarely explicitly taken into account in the models for survival analysis. This study indicates that such a practice is potentially leading to inefficient and biased results, in particular in presence of medium or small sample sizes.     

Adjusting for publication bias: modelling the selection process

RATIONALE, AIMS AND BACKGROUND: Systematic review with meta-analysis, a statistical technique for combining results of several studies, is progressively being used to guide decisions in medicine. Publication bias is acknowledged as a threat to the validity of systematic reviews and its existence may lead to inappropriate decisions about patient management or health policy. It is said to occur when the results of research available in the literature are not representative of the totality of all research. The selection mechanism that causes publication bias is complex, yet despite an extensive literature of empirical research identifying risk factors for publication, little work has been done to improve models of selection. Methods METHODS: that adjust combined meta-analytic estimates for publication bias are compared and applied to a systematic review of oral rehydration solution in the treatment of dehydration. Within a weighted distributions framework models of the selection process are considered and developed further. CONCLUSIONS: Weighted distributions offer a flexible approach that allows the potential to modify the selection function to incorporate other factors. Methods that adjust combined estimates should not be used to provide an alternative answer but to consider the robustness of the combined estimate to publication bias.     

Characteristics of hospital patients receiving medical rehabilitation: an exploratory outcome comparison

This study classified consecutive hospital admissions to acute medical and surgical wards (n = 3,170) into clinical subgroups based on diagnostic, prognostic, and functional criteria. The groups were (1) independent, (2) terminal, (3) medical, (4) dementia, and (5) rehabilitation candidate. Medical record data from hospital admission, discharge, and nine-month follow-up were collected. The groups had unique patterns of survival, residence, and use of health care services during follow-up. Rehabilitation participants were compared with a group similar in age, major diagnostic category, and functional ability, who did not receive rehabilitation. Only 21% of persons meeting the study criteria for rehabilitation actually received rehabilitation services. At follow-up, participants in rehabilitation had lower mortality, spent less time in skilled nursing care, and were less frequently hospitalized. Although exploratory in nature, this study supports previously observed benefits of rehabilitation. In combination with clinical assessment, the process of identifying patient subgroups may be useful in planning interventions more uniformly and in developing measures to reduce selection bias in rehabilitation admission decisions.