medplot: a web application for dynamic summary and analysis of longitudinal medical data based on R

Abstract

In biomedical studies the patients are often evaluated numerous times and a large number of variables are recorded at each time-point. Data entry and manipulation of longitudinal data can be performed using spreadsheet programs, which usually include some data plotting and analysis capabilities and are straightforward to use, but are not designed for the analyses of complex longitudinal data. Specialized statistical software offers more flexibility and capabilities, but first time users with biomedical background often find its use difficult. We developed medplot, an interactive web application that simplifies the exploration and analysis of longitudinal data. The application can be used to summarize, visualize and analyze data by researchers that are not familiar with statistical programs and whose knowledge of statistics is limited. The summary tools produce publication-ready tables and graphs. The analysis tools include features that are seldom available in spreadsheet software, such as correction for multiple testing, repeated measurement analyses and flexible non-linear modeling of the association of the numerical variables with the outcome. medplot is freely available and open source, it has an intuitive graphical user interface (GUI), it is accessible via the Internet and can be used within a web browser, without the need for installing and maintaining programs locally on the user's computer. This paper describes the application and gives detailed examples describing how to use the application on real data from a clinical study including patients with early Lyme borreliosis.

Fig 1. Graphical user interface in the web browser.

The screen is divided in two parts: the sidebar (left part, used to for inputs) and the main panel (right part, used for outputs). The outputs are accessible through various tabs on top of the main panel part of the screen.

Fig 2. Data for first two patients of the demo data set.

Data for two Erythema migrans patients are displayed. It spans eight rows, as each of them was evaluated on four occasions. Not all recorded variables are displayed.

Fig 3. Summary tab output for binary variables.

The table displays the descriptive statistics for the presence of each symptom; the plot shows the observed proportions of patients that report the presence of the symptom, along with their 95% confidence intervals.

Fig 4. Summary: grouping variable tab output for binary variables.

The table displays the summary statistics for the presence of symptoms at baseline for groups defined by the response to treatment at last evaluation. The proportions are compared, unadjusted and adjusted P values and Q values are provided (see text for details).

Fig 5. Graphical exploration tab output for numerical variables—lasagna plot.

The heat map displays graphically the intensity of arthralgia for each patient (horizontal axis) and evaluation occasion (vertical axis). A dendrogram showing patient similarity is plotted on the vertical axis.

Fig 6. Clustering tab output—heat map displaying the similarities of reported symptoms and of patients.

The colors represent the intensity of the symptoms at baseline (rows) for each patient (columns). Hierarchical clustering is used to group symptoms and patients.

Fig 7. Regression model: one evaluation time tab output—estimation of non-linear associations.

The graphs display the estimated associations between the age of the patients and selected symptom intensities at baseline evaluation. Restricted cubic splines are used for modeling. See text for details.