Search

Audit and feedback (Aamp;F) is a common strategy to improve quality of care. Meta-analyses have indicated that Aamp;F may be more effective in realizing desired change when baseline performance is low, it is delivered by a supervisor or colleague, it is provided frequently and in a timely manner, it is delivered in both verbal and written formats, and it includes specific targets and an action plan. However, there is little information to guide operationalization of these factors. Researchers have consequently called for Aamp;F interventions featuring well-described and carefully justified components, with their theoretical rationale made explicit. This paper describes the rationale and development of a quality dashboard including an improvement toolbox for four previous developed pain indicators, guided by Control Theory.

Previously developed quality indicators (QIs) for the intensive care did not give clues for quality improvement. We will improve this by developing new QIs that are actionable, reliable, valid and easy to register. Using a modified RAND technique we developed in three rounds new QIs for blood use at Dutch intensive care units (ICUs). The final set of QIs will be part of the National Intensive Care Evaluation quality registry and evaluated in a randomized controlled trial. In addition, a web-based feedback tool containing strategies for quality improvement will be developed. With the results of this project we hope to improve and maintain good quality of care at the Dutch ICUs .

Feedback to clinicians on their past performance is often aimed at increasing adherence to guidelines. We investigate how various analytical approaches influence the interpretation of adherence data. The analytical approaches vary in considering the actual or the intended use of the feedback, and whether outcomes are inspected over time.At base line, a computerized decision support system was employed at the ICU bedside to increase adherence to a mechanical ventilation strategy. We intervened by providing feedback about adherence to the guideline at the daily ward rounds. The outcome measure was the percentage of ventilation time (VinfT/inf) in excess of the guideline's recommendation. Actual usage of the feedback was logged and data analysis was carried out using two approaches: classical statistics, and statistical process control (SPC) that inspect progress of an outcome over time.Prospective, before/after study.The classical analysis stated that the percentage of ventilation time in excess of the guideline's recommendation decreased significantly due to the feedback (5% reduction, p0.001). When SPC analysis of the outcome was applied, the effect was deemed not significant. When the actual delivery of feedback over time was also included it showed that the experiment does not allow for conclusive results.The concluded effect of providing feedback on adherence to a guideline depends on whether the actual usage pattern of the feedback and the inspection of the outcome over time are considered. Future evaluative studies should report on usage patterns and progression of outcomes over time.

Prognostic models are tools to predict the outcome of disease and disease treatment. These models are traditionally built with supervised machine learning techniques, and consider prognosis as a static, one-shot activity. This paper presents a new type of prognostic model that builds on the Bayesian network methodology that implements a dynamic, process-oriented view on prognosis. In contrast to traditional prognostic models, prognostic Bayesian networks explicate the scenarios that lead to disease outcomes, and can be used to update predictions when new information becomes available. A recursive data analysis strategy for inducing prognostic Bayesian networks from medical data is presented, and applied to data from the field of cardiac surgery. The resulting model outperformed a model that was constructed with off-the-shelf Bayesian network learning software, and had similar performance as class probability trees.

As the support and stabilization of organ function is a major goal of treatment in the Intensive Care Unit (ICU), changes in the function of organ systems are an important indicator of the progression of the disease and recovery. This paper presents how to construct a model that describes changes in organ failure of ICU patients on a day-to-day basis. The model is based on the daily Sequential Organ Failure Assessment (SOFA) scores for six organ systems and predicts, for each of these organ systems, whether failure or recovery is due on the next day, using six logistic regression equations. The joint set of equations, extended with equations for predicting ICU discharge and death, constitutes a firstorder multivariate Markov model. We applied the procedure on a dataset and found that most types of organ failure are highly persistent.

Multiple clinical trials that investigate the same intervention often use different entry criteria to enroll patients into the trials. This variation in entry criteria might lead to the selection of patients with different characteristics. If these characteristics are related to the effect measure (e.g. baseline mortality in a trial investigating mortality reduction) the results of the trials cannot be compared straightforwardly. Therefore it is important to investigate whether the use of different entry criteria leads to the selection of patients with different characteristics.To study differences in the characteristics of the study populations, one could use the data from the control groups of the trials. However, differences in characteristics of these study populations can be caused by differences in entry criteria as well as by clinical diversity. Based on the data from the trials these two causes cannot be distinguished. We propose a method to investigate the influence of differences in entry criteria on the characteristics of study populations. The method corrects for the effects of clinical diversity by simulating the patient selection process on an independent registry database. The method has been successfully applied in the area of severe sepsis.

The importance of terminological systems (TS) to support standardized and structured documentation of medical data is commonly recognized. The usability of TS in real practice strongly depends on the completeness and the correctness of the content of the TS. We here present four different methods that can be applied to evaluate a TS' content. All four methods were applied in a case study. We make a comparison of 1) the results of two methods that focus on the completeness of the content and that differ in the application of the TS that they focus on and 2) the results of an automated and a manual evaluation of the correctness of the content. Finally we summarize the results of all four methods and analyze whether they overlap or complement each other.