Seasonality of Ankle Swelling: Population Symptom Reporting Using Google Trends Fangwei Liu, MD G. Michael Allan, MD Christina Korownyk, MD Michael Kolber, MD, MSc Nigel Flook, MD Harvey Sternberg, MD Scott Garrison, MD, PhD Department of Family Medicine, University of Alberta, Edmonton, Canada
ABSTRACT In our experience, complaints of ankle swelling are more common in summer, typically from patients with no obvious cardiovascular disease. Surprisingly, this observation has never been reported. To objectively establish this phenomenon, we sought evidence of seasonality in the public’s Internet searches for ankle swelling. Our data, obtained from Google Trends, consisted of all related Google searches in the United States from January 4, 2004, to January 26, 2016. Consistent with our expectations and confirmed by similar data for Australia, Internet searches for information on ankle swelling are highly seasonal (highest in midsummer), with seasonality explaining 86% of search volume variability. Ann Fam Med 2016;14:356-358. doi: 10.1370/afm.1953.
INTRODUCTION
E
ach summer we encounter an increased volume of patients complaining of ankle swelling—patients who do not go on to develop cardiovascular, venous, or lymphatic disease. This study looks to provide objective evidence that, for some patients, the symptom burden of ankle swelling is seasonally modulated. Internet search volume has been used successfully to show unexpected seasonality in such conditions as nocturnal leg cramps.1 Using similar methods, we looked for seasonal modulation in the public’s interest in ankle swelling as measured by the volume of Google Internet searches related to ankle swelling.
METHODS
Conflicts of interest: authors report none.
CORRESPONDING AUTHOR
Scott Garrison MD, PhD Evidence Based Medicine Department of Family Medicine 6-60 University Terrace University of Alberta Edmonton Alberta, Canada T6G 2T4 [emailprotected]
The Google Trends search engine provides data from 2004 onward about the frequency with which the public enters specified search terms (www. google.com/trends).2 Reporting can be global or region specific and is presented on a relative scale according to the proportion of overall searches that the queried term represents. We used a composite search term containing any of “ankle swelling,” “swollen ankles,” “swollen feet,” or “swollen legs,” and our data consisted of all such Google searches originating from the United States (the single largest geographic region) during the period of available data (January 4, 2004, to January 26, 2016). We plotted Internet search volume as a time series and performed regressions, using GraphPad Prism 7 (GraphPad Software, Inc), with 2 models that represented either the null hypothesis (no seasonal variation) or the seasonal hypothesis (annual cycling in symptoms). The null model consisted of a best-fit straight line that allowed for linear change in search volume over time but no seasonality. The seasonal model consisted of the best-fit combination of a straight line and a sinusoid according to the following equation: Search_volume = intercept + slope × time + amplitude × sin(2π × time/wavelength + phase shift)
We used a sum-of-squares F test to compare the 2 models and calculate a P value for the difference in fit. We performed an identical analysis for Australia, where seasons are reversed.
A NNALS O F FAMILY MEDICINE
✦
WWW.AN N FA MME D.O R G
356
✦
VO L. 14, N O. 4
✦
J ULY/AUG UST 2016
ANK LE SWELLING
Figure 1. Google Trends Internet search volume for “ankle swelling,” “swollen ankles,” “swollen feet,” or “swollen legs” in the United States from January 4, 2004 to January 26, 2016.
Search volume, % of maximum search observed
100 90 80 70 60 50 40 30 20 10
5
5
01
c2
De
4
01
01
n2
Ju
4
c2
De
3
01
n2
Ju
3
01
c2
De
2
01
01
n2
Ju
2
c2
De
1
01
Ju
n2
1
01
c2
De
01
n2
Ju
01
c2
De
9
Ju
n2
01
9
00
00
c2 De
8
n2
Ju
8
00
c2
De
7
Ju
n2
00
7
00 c2
De
6
00
n2
Ju
6
00
c2
De
5
00
00
n2
Ju
5
c2
De
4
00
n2
Ju
4
00
c2
00
De
c2
00
Ju
De
n2
3
Week of Search Horizontal axis markers (Dec, Jun) are centered mid-month. The dashed line is the best-fit sinusoid.
RESULTS The seasonal model provided a significantly better fit than the null model (P
