Can Timely Outpatient Visits Reduce Readmissions and Mortality Among Heart Failure Patients?

BACKGROUND: Outpatient follow-up after a hospital discharge may reduce the risk of readmissions, but existing evidence has methodological limitations. OBJECTIVES: To assess effect of outpatient follow-up within 7, 14, 21 and 30 days of a hospital discharge on 30-day unplanned readmissions or mortality among heart failure (HF) patients; and whether this varies for patients with different clinical complexities. DESIGN: We analyzed medical records between January 2016 and December 2021 from a prospective cohort study. Using time varying mixed effects parametric survival models, we examined the association between not having an outpatient follow-up and risk of adverse events. We used interaction models to assess if the effect of outpatient follow-up visit on outcomes varies with patients' clinical complexity (comorbidities, grip strength, cognitive impairment and length of inpatient stay). PARTICIPANTS: Two hundred and forty-one patients with advanced HF. MAIN MEASURES: 30-day all-cause (or cardiac) adverse event defined as all cause (or cardiac) unplanned readmissions or death within 30 days of an unplanned all-cause (or cardiac) admission or emergency department visit. KEY RESULTS: We analyzed 1595 all-cause admissions, inclusive of 1266 cardiac admissions. Not having an outpatient follow-up (vs having an outpatient follow-up) significantly increased the risk of 30-day all-cause adverse event. (risk [95% CI] - 14 days: 35.1 [84.5,-1.1]; 21 days: 43.9 [48.2,6.7]; 30 days: 31.1 [48.5, 7.9]) The risk (at 21 days) was higher for those with one co-morbidity (0.25 [0.11,0.58]), mild (0.67 [0.45, 1.00]) and moderate cognitive impairment (0.38 [0.17, 0.84]), normal grip strength (0.57 [0.34, 0.96]) and length of inpatient stay 7-13 days (0.45 [0.23, 0.89]). CONCLUSION: Outpatient follow-up within 30 days after a hospital discharge reduced risk of 30-day adverse events among HF patients, the benefit varying according to clinical complexity. Results suggest the need to prioritize patients who benefit from outpatient follow-up for these visits.

The importance of avoiding time-dependent bias when testing the prognostic value of an intervening event - Two acute cellular rejection examples in intestinal transplantation.

In testing the prognostic value of the occurrence of an intervening event (clinical event that occurs posttransplant), 3 proper statistical methodologies for testing its prognostic value exist (time-dependent covariate, landmark, and semi-Markov modeling methods). However, time-dependent bias has appeared in many clinical reports, whereby the intervening event is statistically treated as a baseline variable (as if it occurred at transplant). Using a single-center cohort of 445 intestinal transplant cases to test the prognostic value of first acute cellular rejection (ACR) and severe (grade of) ACR on the hazard rate of developing graft loss, we demonstrate how the inclusion of such time-dependent bias can lead to severe underestimation of the true hazard ratio (HR). The (statistically more powerful) time-dependent covariate method in Cox's multivariable model yielded significantly unfavorable effects of first ACR (P < .0001; HR = 2.492) and severe ACR (P < .0001; HR = 4.531). In contrast, when using the time-dependent biased approach, multivariable analysis yielded an incorrect conclusion for the prognostic value of first ACR (P = .31, HR = 0.877, 35.2% of 2.492) and a much smaller estimated effect of severe ACR (P = .0008; HR = 1.589; 35.1% of 4.531). In conclusion, this study demonstrates the importance of avoiding time-dependent bias when testing the prognostic value of an intervening event.

Dynamic prediction: A challenge for biostatisticians, but greatly needed by patients, physicians and the public.

Prognosis is usually expressed in terms of the probability that a patient will or will not have experienced an event of interest t years after diagnosis of a disease. This quantity, however, is of little informative value for a patient who is still event-free after a number of years. Such a patient would be much more interested in the conditional probability of being event-free in the upcoming t years, given that he/she did not experience the event in the s years after diagnosis, called "conditional survival." It is the simplest form of a dynamic prediction and can be dealt with using straightforward extensions of standard time-to-event analyses in clinical cohort studies. For a healthy individual, a related problem with further complications is the so-called "age-conditional probability of developing cancer" in the next t years. Here, the competing risk of dying from other diseases has to be taken into account. For both situations, the hazard function provides the central dynamic concept, which can be further extended in a natural way to build dynamic prediction models that incorporate both baseline and time-dependent characteristics. Such models are able to exploit the most current information accumulating over time in order to accurately predict the further course or development of a disease. In this article, the biostatistical challenges as well as the relevance and importance of dynamic prediction are illustrated using studies of multiple myeloma, a hematologic malignancy with a formerly rather poor prognosis which has improved over the last few years.

Metformin and pancreatic cancer survival: Real effect or immortal time bias?

High heterogeneity has been reported among cohort studies investigating the association between metformin and pancreatic cancer survival. Immortal time bias may be one importance source of heterogeneity, as it is widely present in previous cohort studies and may severely impair the validity. Our study aimed to examine whether metformin therapy improves pancreatic cancer survival, and to assess the impact of immortal time bias on the effect estimation of metformin in cohort studies. PubMed, EMbase and SciVerse Scopus were searched. Pooled relative risks (RRs) were derived using a random-effects model. Pooled RR from the six studies without immortal time bias showed no association between metformin and mortality in pancreatic cancer patients (RR 0.93, 95% CI 0.82, 1.05; p = 0.22 and I2 = 75%). In contrast, pooled RR from the nine studies with immortal time bias showed a reduction of 24% in mortality associated with metformin (RR 0.76, 95% CI 0.69, 0.84; p < 0.001 and I2 = 1%). From a meta-regression model, existence of immortal time bias was associated with a reduction of 18% in the effect estimate of metformin on pancreatic cancer survival (ratio of RR 0.82, 95% CI 0.70, 0.96; p = 0.02). In conclusions, cumulative evidence from cohort studies does not support a beneficial effect of metformin on pancreatic cancer survival. The association between metformin and pancreatic cancer survival has been greatly exaggerated in previous cohort studies due to the wide existence of immortal time bias. More rigorous designs and statistical methods are needed to account for immortal time bias.

Exposure density sampling: Dynamic matching with respect to a time-dependent exposure.

Estimating the potential risk associated with an exposure occurring over time requires complex statistical techniques, since ignoring the time from study entry until the exposure leads to potentially seriously biased effect estimates. A prominent example is estimating the effect of hospital-acquired infections on adverse outcomes in patients admitted to the intensive care unit. Exposure density sampling has been proposed as an approach to dynamic matching with respect to a time-dependent exposure. Firstly, exposure density sampling can be useful to reduce the workload of study follow up, as it includes all exposed but only a subset of the not yet exposed individuals. Secondly, it can help to obtain a comparable control group by including propensity score matching. In the present article, we provide the theoretical justification that data obtained by exposure density sampling can be analyzed as a left-truncated cohort. It is shown that exposure density sampling allows estimation of the effect of a time-dependent exposure as well as further baseline covariates on a subsequent event, with only minor loss in precision as compared with a full cohort analysis. The sampling is applied to a real data example (hospital-acquired infections in intensive care units) and in a simulation study. We also provide an estimate of the loss in precision in terms of an increased standard error in the reduced data set after exposure density sampling as compared with the full cohort.

Costs of hospital-acquired Clostridium difficile infections: an analysis on the effect of time-dependent exposures using routine and surveillance data.

BACKGROUND: Hospital-acquired infections have not only gained increasing attention clinically, but also methodologically, as a time-varying exposure. While methods to appropriately estimate extra length of stay (LOS) have been established and are increasingly used in the literature, proper estimation of cost figures has lagged behind. METHODS: Analysing the additional costs and reimbursements of Clostridium difficile-infections (CDI), we use a within-main-diagnosis-time-to-exposure stratification approach to incorporate time-varying exposures in a regression model, while at the same time accounting for cost clustering within diagnosis groups. RESULTS: We find that CDI is associated with €9000 of extra costs, €7800 of higher reimbursements, and 6.4 days extra length of stay. Using a conventional method, which suffers from time-dependent bias, we derive estimates more than three times as high (€23,000, €8000, 21 days respectively). We discuss our method in the context of recent methodological advances in the estimation of the costs of hospital-acquired infections. CONCLUSIONS: CDI is associated with sizeable in-hospital costs. Neglecting the methodological particularities of hospital-acquired infections can however substantially bias results. As the data needed for an appropriate analysis are collected routinely in most hospitals, we recommend our approach as a feasible way for estimating the economic impact of time-varying adverse events during hospital stay.

Multistate methodology improves risk assessment under time-varying drug intake-a new view on pregnancy outcomes following coumarin exposure.

PURPOSE: Observational cohort studies are essential to evaluate the risk of adverse pregnancy outcomes associated with drug intake. Besides left truncation and competing events, it is crucial to account for the time-dynamic pattern of drug exposure. In fact, potentially harmful medications are often discontinued, which might affect the outcome. Ignoring these challenges may lead to biased estimation of drug-related risks highlighting the need for adequate statistical techniques. METHODS: We reanalyze updated data of a recently published study provided by the German Embryotox pharmacovigilance institute. The aim of the study was to quantify the effect of discontinuation of vitamin K antagonist phenprocoumon on the risk of spontaneous abortion. RESULTS: We outline multistate methodology as a powerful method removing bias in probability estimation inherent to commonly used crude proportions. We incorporate time-dependent discontinuation and competing pregnancy outcomes as separate states in a multistate model, which enables the formulation of hazard-based Cox proportional hazard models and the application of so-called landmark techniques. Results show that early discontinuation of phenprocoumon substantially reduces the risk of spontaneous abortion, which is of great importance for both pregnant women and treating physicians. CONCLUSIONS: An adequate handling of discontinuation times is essential when analyzing the risk of spontaneous abortion. The proposed concepts are not restricted to pregnancy outcome studies but have broad usage in other fields of epidemiology. Our nontechnical report may provide guidance for the design and analysis of future studies. Example code is provided.

A Time-varying Covariate Approach for Survival Analysis of Paediatric Outcomes.

BACKGROUND: Conventional survival analysis is commonly applied in the analysis of time-to-event data in paediatric studies, where the exposure variables of interest are often treated as time-fixed. However, the values of these exposure variables can vary over time and time-fixed analysis may introduce time-dependent bias. METHODS: Time-dependent bias is illustrated graphically considering two scenarios in longitudinal study settings for paediatric time-to-event outcomes. As an illustrative example, the time-varying covariate approach was applied to survival analysis of breast-feeding data (n = 695) collected in China between 2010 and 2011, with an emphasis on the effects of covariates 'solid foods introduction' and 'maternal return to work' on breast-feeding duration up to 12 months postpartum. RESULTS: Time-varying exposures could occur before or after the occurrence of an event of interest so that time-fixed analysis can lead to biased and imprecise parameter estimates. In the illustrative example, the reduced risk of 'solid foods introduction' (hazard ratio (HR) 0.61, 95% confidence interval (CI) 0.50, 0.75) on breast-feeding cessation and an absence of an association with 'maternal return to work' (HR 0.99, 95% CI 0.73, 1.36) from the time-fixed analysis reversed (HR 1.50, 95% CI 1.17, 1.93) and became significant (HR 1.45, 95% CI 1.06, 2.00), respectively, based on the time-varying covariate model. CONCLUSIONS: The time-varying covariate approach is preferable for survival analysis of time-to-event data in the presence of time-varying exposures.

Statistical Analysis of Clinical COVID-19 Data: A Concise Overview of Lessons Learned, Common Errors and How to Avoid Them.

By definition, in-hospital patient data are restricted to the time between hospital admission and discharge (alive or dead). For hospitalised cases of COVID-19, a number of events during hospitalization are of interest regarding the influence of risk factors on the likelihood of experiencing these events. The same is true for predicting times from hospital admission of COVID-19 patients to intensive care or from start of ventilation (invasive or non-invasive) to extubation. This logical restriction of the data to the period of hospitalisation is associated with a substantial risk that inappropriate methods are used for analysis. Here, we briefly discuss the most common types of bias which can occur when analysing in-hospital COVID-19 data.

Estimating excess length of stay due to healthcare-associated infections: a systematic review and meta-analysis of statistical methodology.

BACKGROUND: Healthcare-associated infection (HCAI) affects millions of patients worldwide. HCAI is associated with increased healthcare costs, owing primarily to increased hospital length of stay (LOS) but calculating these costs is complicated due to time-dependent bias. Accurate estimation of excess LOS due to HCAI is essential to ensure that we invest in cost-effective infection prevention and control (IPC) measures. AIM: To identify and review the main statistical methods that have been employed to estimate differential LOS between patients with, and without, HCAI; to highlight and discuss potential biases of all statistical approaches. METHODS: A systematic review from 1997 to April 2017 was conducted in PubMed, CINAHL, ProQuest and EconLit databases. Studies were quality-assessed using an adapted Newcastle-Ottawa Scale (NOS). Methods were categorized as time-fixed or time-varying, with the former exhibiting time-dependent bias. Two examples of meta-analysis were used to illustrate how estimates of excess LOS differ between different studies. FINDINGS: Ninety-two studies with estimates on excess LOS were identified. The majority of articles employed time-fixed methods (75%). Studies using time-varying methods are of higher quality according to NOS. Studies using time-fixed methods overestimate additional LOS attributable to HCAI. Undertaking meta-analysis is challenging due to a variety of study designs and reporting styles. Study differences are further magnified by heterogeneous populations, case definitions, causative organisms, and susceptibilities. CONCLUSION: Methodologies have evolved over the last 20 years but there is still a significant body of evidence reliant upon time-fixed methods. Robust estimates are required to inform investment in cost-effective IPC interventions.

Survival biases lead to flawed conclusions in observational treatment studies of influenza patients.

BACKGROUND AND OBJECTIVE: Several observational studies reported that Oseltamivir (Tamiflu) reduced mortality in infected and hospitalized patients. Because of the restriction of observation to hospital stay and time-dependent treatment assignment, such findings were prone to common types of survival bias (length, time-dependent and competing risk bias). METHODS: British hospital data from the Influenza Clinical Information Network (FLU-CIN) study group were used which included 1,391 patients with confirmed pandemic influenza A/H1N1 2009 infection. We used a multistate model approach with following states: hospital admission, Oseltamivir treatment, discharge, and death. Time origin is influenza onset. We displayed individual data, risk sets, hazards, and probabilities from multistate models to study the impact of these three common survival biases. RESULTS: The correct hazard ratio of Oseltamivir for death was 1.03 (95% confidence interval [CI]: 0.64-1.66) and for discharge 1.89 (95% CI: 1.65-2.16). Length bias increased both hazard ratios (HRs): HR (death) = 1.82 (95% CI: 1.12-2.98) and HR (discharge) = 4.44 (95% CI: 3.90-5.05), whereas the time-dependent bias reduced them: HR (death) = 0.62 (95% CI: 0.39-1.00) and HR (discharge) = 0.85 (95% CI: 0.75-0.97). Length and time-dependent bias were less pronounced in terms of probabilities. Ignoring discharge as a competing event for hospital death led to a remarkable overestimation of hospital mortality and failed to detect the reducing effect of Oseltamivir on hospital stay. CONCLUSIONS: The impact of each of the three survival biases was remarkable, and it can make neuraminidase inhibitors appear more effective or even harmful. Incorrect and misclassified risk sets were the primary sources of biased hazard rates.

The prolongation of length of stay because of Clostridium difficile infection.

BACKGROUND: Clostridium difficile infection (CDI) possibly extends hospital length of stay (LOS); however, the current evidence does not account for the time-dependent bias, ie, when infection is incorrectly analyzed as a baseline covariate. The aim of this study was to determine whether CDI increases LOS after managing this bias. METHODS: We examined the estimated extra LOS because of CDI using a multistate model. Data from all persons hospitalized >48 hours over 4 years in a tertiary hospital in Australia were analyzed. Persons with health care-associated CDIs were identified. Cox proportional hazards models were applied together with multistate modeling. RESULTS: One hundred fifty-eight of 58,942 admissions examined had CDI. The mean extra LOS because of infection was 0.9 days (95% confidence interval: -1.8 to 3.6 days, P = .51) when a multistate model was applied. The hazard of discharge was lower in persons who had CDI (adjusted hazard ratio, 0.42; P < .001) when a Cox proportional hazard model was applied. CONCLUSION: This study is the first to use multistate models to determine the extra LOS because of CDI. Results suggest CDI does not significantly contribute to hospital LOS, contradicting findings published elsewhere. Conversely, when methods prone to result in time-dependent bias were applied to the data, the hazard of discharge significantly increased. These findings contribute to discussion on methods used to evaluate LOS and health care-associated infections.

Hospital-acquired infections--appropriate statistical treatment is urgently needed!

Research on hospital-acquired infections (HAIs) requires the highest methodological standards to minimize the risk of bias and to avoid misleading interpretation. There are two major issues related specifically to studies in this area, namely the timing of infection and the occurrence of so-called competing risks, which deserve special attention. Just as a patient who acquires a serious infection during hospital admission needs appropriate antibiotic treatment, data being collected in studies on hospital-acquired infections need appropriate statistical analysis. We illustrate the urgent need for appropriate statistical treatment of hospital-acquired infections with some examples from recently conducted studies.The considerations presented are relevant for investigations on risk factors for HAIs as well as for outcome studies.