To The Editor:
Interpreting repeated measures in longitudinal data analysis can be a challenge, especially in large studies and in studies with long follow-up periods. Our research team from the Breast cancer in Older Women study1,2 struggled to figure out how to visually depict data on surveillance mammograms among a cohort of breast cancer survivors—with the overall goal of trying to determine whether surveillance mammography in older women is associated with decreased breast cancer mortality. We have long-term follow-up data on surveillance mammography among a cohort of 1859 breast cancer survivors. To make things more challenging, our outcome (breast cancer mortality) is dichotomous and follow-up is censored by other events (ie, breast cancer recurrence, mortality due to other cause, and disenrollment). With over 1800 women and 15 years of follow-up, spaghetti plots were incomprehensible.
It was with great interest that we came across the proposal by Swihart and colleagues3 in EPIDEMIOLOGY who proposed a new type of graph—the lasagna plot. The authors nicely explained the advantages of lasagna plots over spaghetti plots to display longitudinal data. Lasagna plots incorporate time (horizontal axis), values for repeated measures (colors or shadings), and space for displaying each subject (vertical axis) without overlapping data. Moreover, through various sorting and clustering strategies, lasagna plots make it easier to visually detect patterns of repeated measures in data.
In the article of Swihart et al,3 R code was provided to produce lasagna plots.4 However, our research team was not sufficiently versed in R to manipulate the code for our purposes. We contacted Dr. Swihart to find out if there was a code to create lasagna plots in other statistical packages. However, there was not, and so we set out to recreate these plots with SAS software (SAS Institute Inc., Cary, NC).5 We found that lasagna plot can be regarded as a nongeographic map using the GMAP procedure in SAS/GRAPH software. A grid is created as a “map”; values of repeated measures for each subject (represented by a specific color) are filled in for each corresponding unit on the grid; and labels and reference lines are added in the last step by annotations.
We duplicated the lasagna plots in Dr. Swihart’s commentary, and Dr. Swihart replicated the plots from their previously published data using our SAS code. The SAS code and the corresponding lasagna plots are available in an eAppendix (https://links.lww.com/EDE/A616). We are happy to provide this alternative baking option for lasagna plots for people like us with different kitchens.
Hongyuan Gao
Diana S. M. Buist
Group Health Research Institute
Seattle, WA
[email protected]
Timothy L. Lash
Department of Epidemiology & Prevention
Division of Public Health Sciences
Wake Forest School of Medicine
Department of Clinical Epidemiology
Aarhus University
Aarhus, Denmark
Jaclyn L. F. Bosco
Dana-Farber Cancer Institute
Boston Medical Center
Boston, MA
Bruce Swihart
Department of Biostatistics
Johns Hopkins Bloomberg School of
Public Health
Baltimore, MD
REFERENCES
1. Enger SM, Thwin SS, Buist DS, et al. Breast cancer treatment of older women in integrated health care settings. J Clin Oncol. 2006;24:4377–4383
2. Lash TL, Fox MP, Buist DS, et al. Mammography surveillance and mortality in older breast cancer survivors. J Clin Oncol. 2007;25:3001–3006
3. Swihart BJ, Caffo B, James BD, Strand M, Schwartz BS, Punjabi NM. Lasagna plots: a saucy alternative to spaghetti plots. Epidemiology. 2010;21:621–625
4. Swihart BJ, Caffo B, James JD, et al. eAppendix 1: Lasagne plots: a saucy alternative to spaghetti plots.
https://links.lww.com/EDE/A401
5. SAS Institute Inc. SAS/GRAPH 9.2 Reference. 20102nd ed Cary, NC SAS Institute Inc