The retrospective chart review: important methodological considerations.
Vassar M, Holzmann M. · 2013
Read The retrospective chart review: important methodological considerations. by Vassar M, Holzmann M. (2013) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/europmc-pmc3853868.txt.
Introduction
Matt Vassar and Matthew Holzmann review chart review methods in the Journal of Educational Evaluation for Health Professions in 2013. A retrospective chart review mines patient records that already exist to answer research questions. It is cheap and wide in reach, yet easy to do badly.
The authors list ten common method mistakes in such reviews. They span vague questions, sloppy record picking, untrained abstractors, and missing checks on data quality. For most flaws they give fixes or point to guides that help. Their aim is a best-practice path for planning, running, and judging chart reviews, so the cheap design yields results that later, costlier studies can build on with trust.
Catalog record
| CREATOR | Vassar M, Holzmann M. |
|---|---|
| YEAR | 2013 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 32 KB |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
The retrospective chart review: important methodological considerations By Vassar Matt, Holzmann Matthew Published in: Journal of Educational Evaluation for Health Professions DOI: 10.3352/jeehp.2013.10.12 ABSTRACT Abstract In this paper, we review and discuss ten common methodological mistakes found in retrospective chart reviews. The retrospective chart review is a widely applicable research methodology that can be used by healthcare disciplines as a means to direct subsequent prospective investigations. In many cases in this review, we have also provided suggestions or accessible resources that researchers can apply as a “best practices” guide when planning, conducting, or reviewing this investigative method. Keywords: Best practices, Chart review, Medical record, Methodology, Retrospective review INTRODUCTION The retrospective chart review (RCR), also known as a medical record review, is a type of research design in which pre-recorded, patient-centered data are used to answer one or more research questions [1]. The data used in such reviews exist in many forms: electronic databases, results from diagnostic tests, and notes from health service providers to mention a few. RCR is a popular methodology widely applied in many healthcare-based disciplines such as epidemiology, quality assessment, professional education and residency training, inpatient care, and clinical research (cf. Gearing et al. [2]), and valuable information may be gathered from study results to direct subsequent prospective studies. Attesting to the popularity of this technique, a review of three emergency medicine journals revealed that nearly one-quarter of all research published within the study’s timeframe used RCR methodology [3]. Gilbert et al. [3] also examined methodological rigor in reporting practices from the RCRs in their sample. Predictably, they found that the majority of studies lacked sound methodological standards. Since poor methodology is a principal reason for peer-reviewed journal rejections, the aim of this article is to discuss common methodological mistakes and omissions made when conducting RCRs. The following methodological points stem from personal consultation experience as well as from the works of Gilbert et al. [3], Gearing et al. [2], Worster and Haines [1], and Findley and Daum [4]–all of which provide valuable information to consider when planning an RCR. When formulating the ideas presented herein, we used the standards provided by Gilbert et al. [3] to structure this paper while incorporating additional considerations which we deem to be important. COMMON MISTAKES WHEN CONDUCTING A RETROSPECTIVE CHART REVIEW 1. Failure to create well-defined, clearly-articulated research questions The first step when planning a RCR is to formulate a series of research questions that are to be answered based on results of the study. Research questions should be logical, flowing from that which is known or believed to be true to that which is unknown and requires validation [5]. Research questions form the initial structure of the RCR and guide the study design and data analysis. It is important to spend adequate time carefully scripting and revising the research questions for the study. There is no shortage of published advice on developing and refining research questions . We have selected one framework for the design and articulation of research questions to present here which we have found to be particularly useful. Though not mentioned elsewhere in this manuscript, we recommend Morgan and Harmon [6] to the reader as an additional reference in properly framing research questions. The framework presented here is a typology of research questions. Such questions generally fall into one of three categories: questions of description, questions of relationship, or questions of comparison. Questions of description are common to RCRs. These questions describe what is going on or what exists [7]. Incidence and prevalence research are descriptive. For example, we might formulate the research question, “What is the incidence rate for seasonal influenza among the elderly population in Belgium for the year 2009?” The answer to this question might be expressed in terms of a percentage. Other examples might include questions comparing characteristics and outcomes of patients with community-acquired pneumonia who were admitted to the Intensive Care Unit (ICU) with those patients managed on the ward [8] or characterizing hip joint pain referral patterns [9]. Results to descriptive questions are often reported as proportions, percentages, frequency counts, measures of central tendency (mean, median, mode), measures of variability (standard deviation, range), or various charts, graphs, and tables. Questions of relationship ask how phenomena are related to one another [10]. As an example, we might pose the question, “What is the relationship between occupational burnout and suicide ideation among medical residents in the Northeast United States?” To answer this question, we would likely gather burnout and suicide ideation scores from the population of interest and then calculate a correlation coefficient to quantify this relationship. Other examples of this type might include examining the relationship between levels of community-reported infectious diseases and rate of neural tube defects [11] or assessing the relationship between the use of antipseudomonal drugs and the development of a resistance to Pseudomonas aeruginosa [12]. The answers to these types of questions are often provided in the form of a correlation coefficient. There are many such coefficients, and the proper choice of the coefficient is dictated by the nature of the data, including data level (nominal, ordinal, interval, or ratio) and the underlying distribution. Questions of comparison ask about group or sub-group differences on a variable (or variables) of interest. The groups discussed in the above definition represent levels of the independent variable, whereas the variable examined across groups is known as the dependent variable. Questions of comparison are often used in randomized clinical trials. In a simple example, a group of patients with a particular disorder are randomly assigned to either a treatment or to a control group. The treatment group receives the intervention while the control group does not. At the end of the trial, the two groups are compared to assess the efficacy of the treatment. While questions of comparison may seek to establish cause-effect relationships, such is not always the case. We might pose the research question, “Are there differences between males and females on life satisfaction following a spinal cord injury?” In this example, the independent variable cannot be randomly assigned since gender is a pre-determined characteristic. This question still lends itself to comparison however. Other examples might include comparing the effect of fluid resuscitation with albumin or saline on mortality among ICU patients [13] or comparing four weight loss diets from low to high carbohydrate intake for effects on weight loss [14]. These types of questions are often answered by statistically comparing measures of central tendency across groups. 2. Failure to consider sampling issues a priori There are two main issues that need to be addressed with respect to sampling considerations: the sample size and the sampling strategy. A mathematical process called power analysis can be used to help determine the number of charts needed for a particular study. Power refers to the probability that a statistical test will reject the null hypothesis when the alternative hypothesis is true. Let us consider an example to illustrate power more clearly. In a previously mentioned example, we posed a research question related to gender differences in life satisfaction following a spinal cord injury. The null hypothesis is always stated to reflect no difference. In this example, the null hypothesis would state that no