Scaled rectangle diagrams can be used to visualize clinical and epidemiological data

doi:10.1016/j.jclinepi.2005.01.018

Journal of Clinical Epidemiology

Volume 58, Issue 10, October 2005, Pages 974-981

https://doi.org/10.1016/j.jclinepi.2005.01.018 Get rights and content

Abstract

Objective

To illustrate scaled rectangle diagrams as a method for displaying clinical and epidemiological attributes (such as symptoms, signs, results of marker tests, disease, or risk factors). These are quantitative Venn diagrams, but using rectangles instead of circles.

Study Design and Setting

The method is illustrated through examples from various data sets with different types of clinical information.

Results

Examples drawing on studies of lung disease, rheumatic fever, blood pressure, lipid levels, sudden infant death syndrome, and low birth weight illustrate the different types of relationships between variables that the scaled rectangle approach can reveal (e.g., high- and low-risk groups; dependent, independent, or co-occurring attributes; effects from choice of cutoff; cumulative distributions; and case-control attributes).

Conclusion

Scaled rectangle diagrams are a novel way to display clinical data. They show clearly the relative frequency of clinical attributes and the extent to which they are shared characteristics. Features are revealed that might otherwise not have been appreciated.

Introduction

Venn diagrams have often been suggested in clinical research to symbolically show how features such as clinical signs, symptoms, tests, and staging coincide. Feinstein [1] pioneered their use. Some authors have tried scaling Venn diagrams so that the circles and their intersections are drawn in proportion to frequencies of occurrence. Feinstein [1], however, noted the technical difficulty in creating such proportional Venn diagrams and observed that, although they may demonstrate quantitative relationships, the qualitative aspects of the relationships may be less clear than with a purely symbolic nonproportional Venn diagram. In an earlier publication [2], I have shown that the technical difficulty is reduced somewhat by using rectangles instead of circles. The resulting scaled rectangle diagram often show both qualitative and quantitative information reasonably well.

Here, examples of the analysis of clinical and epidemiological data will illustrate the use of scaled rectangle diagrams. With examples used as case studies, some of the ideas I had earlier presented [2] will be advanced in a number of ways. For example, despite my original suggestion that four attributes may represent a limit [2], it is sometimes possible to represent more. New criteria for fitting scaled rectangle diagrams use a penalizing function to avoid narrow rectangles with a high length-to-breadth ratio. Color or shading can be used to represent the intensity of an additional variable, and scaled rectangle diagrams can be used to represent combinations of attributes and high- and low-risk subgroups. In some examples, scaled rectangle diagrams will be compared with Feinstein's original Venn diagrams. Other examples draw on studies of hypertension, lipids, sudden infant death syndrome, and low birth weight to illustrate the method.

Section snippets

Theory and methods

Suppose there exist attributes (e.g., signs, symptoms, or tests) that are measurable on patients and there is a data set with recorded presence of attributes in a sample of patients. The relative frequency, or prevalence, of each attribute can be computed and rectangles drawn with area proportional to prevalence. A scaled rectangle diagram is formed by positioning these rectangles so that not only are the areas of the rectangles proportional to attribute prevalence, but also with areas of

Feinstein's lung disease example

Feinstein [1] describes a sample of 175 patients with lung disease. Patients were classified with the characteristics emphysema (EMPH), mucous gland hyperplasia (MPH), ‘blue and bloated’ (BB), ‘pink and puffy’ (PP), and with neither BB nor PP (NN). Feinstein presented two attempts to draw a quantitative Venn diagram; the better of these is reproduced in Fig. 1a. Figure 1b is a scaled rectangle diagram for these data. It shows the relative sizes of the subgroups more accurately than Feinstein's

Discussion

Clinical and epidemiological researchers are often interested in the extent to which signs, symptoms, results of tests, risk factors, and other clinical observations co-occur. Often an appreciation of the actual extent of co-occurrence is lost by the reduction to summary measures of association, which are usually intended to convey departure from statistical independence; however, although independence is a fundamental concept, it is instructive to actually visualize the co-occurrence of

References (16)

K.M. Johnson
The two by two diagram: a graphical truth table
J Clin Epidemiol
(1999)
J.J. Foster
The influence of shape on apparent area: a new demonstration
Acta Psychol
(1976)
A.R. Feinstein
Clinical judgment
(1967)
R.J. Marshall
Displaying clinical data relationships with scaled rectangle diagrams
Stat Med
(2001)
Marshall RJ. Search Partition Analysis. SPAN software home page [Internet]. 1999–2001. Available at:...
C. Bullen et al.
Cardiovascular disease risk factors in 65–84 year old men and women: results from the Auckland University Heart and Health Study
N Z Med J
(1998)
E.A. Mitchell et al.
Results from the first year of the New Zealand cot death study
N Z Med J
(1991)
R.J. Marshall
Determining and visualising at-risk groups in case-control data
J Epidemiol Biostat
(2001)

There are more references available in the full text version of this article.

Cited by (27)

Physiotherapy for patients with shoulder pain in primary care: a descriptive study of diagnostic- and therapeutic management
2017, Physiotherapy (United Kingdom)
Citation Excerpt :
In all other cases, median scores and the interquartile range (IQR) were used. Hypotheses after patient history were categorized according to the guidelines (complaints arising from pathology/dysfunction in: 1) the subacromial space (subacromial impingement, internal impingement & sprain/strain), 2) glenohumeral joint (glenohumeral joint instability, frozen shoulder, biceps tendinopathy & SLAP), 3) acromioclavicular (AC)/sternoclavicular (SC) joint, 4) cervico-thoracic spine and 5) other and presented in a scaled rectangle diagram [30]. The number of missings were reported with all data.
Shoulder pain is one of the three main musculoskeletal complaints and more than 50% of the patients have symptoms longer than 6 months. Until now, limited data exist about the content of physiotherapy for patients with shoulder pain in primary care.
Describe current physiotherapeutic diagnostic- and therapeutic management, including the use of diagnostic ultrasound, in patients with shoulder pain in primary care.
A prospective cohort study in primary care physiotherapy with a 12 week follow-up.
Descriptive data from physiotherapists was collected, such as: the diagnostic hypotheses after patient history and physical examination, the use of specific tests and diagnostic ultrasound, the interventions used and possible changes in treatment plan.
Subacromial impingement syndrome was the most common hypothesis after patient history (48%) as well as physical examination (39%). Diagnostic ultrasound was used in 31% and of these patients the clinical diagnosis changed in 29%. Various interventions were used in all clinical diagnoses. After 12 weeks 41% of patients still received physiotherapy treatment.
Patients with shoulder pain in physiotherapy practice frequently show signs of subacromial impingement syndrome. The interventions used by the physiotherapists were generally in line with the guideline for subacromial impingement syndrome however a small proportion of physiotherapists used massage and tape/bracing techniques. A large proportion of patients were still receiving treatment after 12 weeks when no improvement was observed. If treatment for patients with subacromial impingement shows no benefit patients should be referred back to the general practitioner or orthopedic surgeon. Conclusions from this study might be slightly biased because of the selection of physiotherapists.
XTranscription factor binding in human cells occurs in dense clusters formed around cohesin anchor sites
2013, Cell
During cell division, transcription factors (TFs) are removed from chromatin twice, during DNA synthesis and during condensation of chromosomes. How TFs can efficiently find their sites following these stages has been unclear. Here, we have analyzed the binding pattern of expressed TFs in human colorectal cancer cells. We find that binding of TFs is highly clustered and that the clusters are enriched in binding motifs for several major TF classes. Strikingly, almost all clusters are formed around cohesin, and loss of cohesin decreases both DNA accessibility and binding of TFs to clusters. We show that cohesin remains bound in S phase, holding the nascent sister chromatids together at the TF cluster sites. Furthermore, cohesin remains bound to the cluster sites when TFs are evicted in early M phase. These results suggest that cohesin-binding functions as a cellular memory that promotes re-establishment of TF clusters after DNA replication and chromatin condensation.
Cardiovascular risk can be represented by scaled rectangle diagrams
2009, Journal of Clinical Epidemiology
Citation Excerpt :
Here the nature of the difficulty is demonstrated with “scaled rectangle diagrams,” that in contrast to the approach in Ref. [1], represent frequencies by areas. Scaled rectangle diagrams have been developed to show relationships among categorical variables and specifically in an epidemiologic and clinical context [2]. We use data from a CVD screening program in Auckland, New Zealand, from published Scottish data, and data from a modeling study of CVD prevention in China, to illustrate the ideas.
To show how levels of cardiovascular disease (CVD) risk and occurrence of CVD can be graphically represented.
Methodological and using graphics to illustrate CVD risk prediction. Uses data on CVD risk and outcomes in three studies from New Zealand, China, and Scotland with levels of risk determined by multivariate risk scores.
The diagrams clearly show that risk levels and actual CVD occurrence do not often occur together, many individuals at high risk do not experience a CVD event and conversely, many CVD events do not occur in the labeled high risk.
Scaled rectangles can easily be formed to show how modeled predicted risk levels and actual occurrence of CVD coincide. Poor discrimination is well illustrated.
Clinical information displays to improve ICU outcomes
2008, International Journal of Medical Informatics
In a previous study, we compared a prototype ecological display (ED) that represented physiological data in a structured pictorial format with two bar graph displays [J.A. Effken, Improving clinical decision making through ecological interfaces, Ecol. Psych. 18 (2006) 283–318]. In ED and the first bar graph display, data were grouped hierarchically based on a cognitive work analysis (CWA); in the second bar graph display they were grouped as usually collected. Treatment efficiency (i.e., percentage of time seven variables in the CWA model were in target range) improved similarly with the two displays incorporating the CWA order for intensive care unit (ICU) residents, but not for novice ICU nurses. Hypothesized reasons for this result included: insufficient practice with novel displays; use of identical histories across displays; insufficient clinical knowledge; and the variables used in the efficiency analysis, which included only one of ED's four integrated design elements. In the current study we tested these hypotheses.
We asked ICU nurses assigned to three knowledge groups based on intensive care and hemodynamic monitoring pretests to identify and treat oxygenation problems presented via ED and the first bar graph display (BGD) in an experimental laboratory simulation. We measured the impact of display, clinical scenario, data level, knowledge, presentation order, and practice extent on event recognition, treatment efficiency, cognitive workload, and user satisfaction.
The two displays produced little difference in recognition speed or overall cognitive workload, but user satisfaction was greater with ED. When 12 variables were included in the analysis, treatment efficiency improved with ED; when only 7 were measured, BGD prevailed. The results suggest benefits for the kind of synthesis provided in ED, but also a potential limitation. If too many different pictorial formats are used in a display, detecting critical events may be more difficult.
Individual and Joint Expert Judgments as Reference Standards in Artifact Detection
2008, Journal of the American Medical Informatics Association
Citation Excerpt :
Furthermore, the table shows that high agreement values were observed for no pair of physicians for all three variables. Figure 3 visualizes the intersection of the individual and joint judgments of the ABPm, CVP, and HR time series using scaled rectangle diagrams.18 The CVP diagram shows that almost all data points that were judged as artifact by expert 1 and 2 were included in the data points regarded as artifact by expert 3 and 4.
To investigate the agreement among clinical experts in their judgments of monitoring data with respect to artifacts, and to examine the effect of reference standards that consist of individual and joint expert judgments on the performance of artifact filters.
Individual judgments of four physicians, a majority vote judgment, and a consensus judgment were obtained for 30 time series of three monitoring variables: mean arterial blood pressure (ABPm), central venous pressure (CVP), and heart rate (HR). The individual and joint judgments were used to tune three existing automated filtering methods and to evaluate the performance of the resulting filters.
The interrater agreement was calculated in terms of positive specific agreement (PSA). The performance of the artifact filters was quantified in terms of sensitivity and positive predictive value (PPV).
PSA values between 0.33 and 0.85 were observed among clinical experts in their selection of artifacts, with relatively high values for CVP data. Artifact filters developed using judgments of individual experts were found to moderately generalize to new time series and other experts; sensitivity values ranged from 0.40 to 0.60 for ABPm and HR filters (PPV: 0.57–0.84), and from 0.63 to 0.80 for CVP filters (PPV: 0.71–0.86). A higher performance value for the filters was found for the three variable types when joint judgments were used for tuning the filtering methods.
Given the disagreement among experts in their individual judgment of monitoring data with respect to artifacts, the use of joint reference standards obtained from multiple experts is recommended for development of automatic artifact filters.
Proportional classifications of COPD phenotypes
2008, Thorax
Chronic obstructive pulmonary disease (COPD) encompasses a group of disorders characterised by the presence of incompletely reversible airflow obstruction with overlapping subsets of different phenotypes including chronic bronchitis, emphysema or asthma. The aim of this study was to determine the proportion of adult subjects aged >50 years within each phenotypic subgroup of COPD, defined as a post-bronchodilator ratio of forced expiratory volume in 1 s/forced vital capacity (FEV₁/FVC) <0.7, in accordance with current international guidelines.
Adults aged >50 years derived from a random population-based survey undertook detailed questionnaires, pulmonary function tests and chest CT scans. The proportion of subjects in each of 16 distinct phenotypes was determined based on combinations of chronic bronchitis, emphysema and asthma, with and without incompletely reversible airflow obstruction defined by a post-bronchodilator FEV₁/FVC ratio of 0.7.
A total of 469 subjects completed the investigative modules, 96 of whom (20.5%) had COPD. Diagrams were constructed to demonstrate the relative proportions of the phenotypic subgroups in subjects with and without COPD. 18/96 subjects with COPD (19%) had the classical phenotypes of chronic bronchitis and/or emphysema but no asthma; asthma was the predominant COPD phenotype, being present in 53/96 (55%). When COPD was defined as a post-bronchodilator FEV₁/FVC less than the lower limit of normal, there were one-third fewer subjects with COPD and a smaller proportion without a defined emphysema, chronic bronchitis or asthma phenotype.
This study provides proportional classifications of the phenotypic subgroups of COPD which can be used as the basis for further research into the pathogenesis and treatment of this heterogeneous disorder.

View all citing articles on Scopus

View full text

Original ArticleScaled rectangle diagrams can be used to visualize clinical and epidemiological data

Abstract

Objective

Study Design and Setting

Results

Conclusion

Introduction

Section snippets

Theory and methods

Feinstein's lung disease example

Discussion

J Clin Epidemiol

Acta Psychol

Clinical judgment

Displaying clinical data relationships with scaled rectangle diagrams

Stat Med

Cardiovascular disease risk factors in 65–84 year old men and women: results from the Auckland University Heart and Health Study

N Z Med J

Results from the first year of the New Zealand cot death study

N Z Med J

Determining and visualising at-risk groups in case-control data

J Epidemiol Biostat

Original Article
Scaled rectangle diagrams can be used to visualize clinical and epidemiological data