#Preliminaries:
knitr::opts_chunk$set(message=FALSE, warning=FALSE)

rm(list=ls())

library(tidyverse)
library(lubridate)
library(viridis)
#library(devtools)
#devtools::install_github("PMassicotte/gtrendsR") # installation from github necessary
library(gtrendsR)
library(forecast)
library(jtools)
library(sandwich)

theme_set(theme_light())

dir.create("filesTrends", showWarnings = FALSE)

Introduction

Back pain is a common global health problem, driven by an ageing population and insufficient or inappropriate physical activity. In addition, from 2020 onward, physical exercise decreased strongly due to policies implemented to mitigate the COVID-19 pandemic. Furthermore, while working from home became the norm during this time, many people lacked ergonomic setups, potentially further intensifying the issue. Hence, a natural question arises: did the political lockdowns negatively influence the prevalence of back pain?

But how can changes in back pain over time be detected? One approach is to use Google Trends, which provides quasi-instant access to aggregated queries from Google users. For example, Hoerger et al. (2020), Knipe et al. (2020), and Brodeur et al. (2021) used Google Trends to study the impact of the pandemic on mental health and well-being. Szilagyi et al. (2021) compared queries for back pain, before and after the pandemic, while Ciaffi et al. (2021) analyzed the seasonality of back pain in Italy. In the following analysis, we see how health issues can be traced back to the COVID-19 pandemic by combining Google Trends data from Germany with statistical methods.

Data Analysis

Let us examine the relative frequency of internet queries for back pain-related keywords from 2016 onward. We focus on queries from Germany and assume that the back pain synonyms ‘Rückenschmerzen’, ‘Rücken Schmerzen’, ‘Rückenschmerz’, and ‘Rücken Schmerz’ reflect overall interest in this topic, serving as a proxy for the associated burden of disease over time. The data is provided on a monthly level.

#trends <- gtrends(keyword = c("Schmerz Rücken + Rückenschmerz + Schmerzen Rücken + Rückenschmerzen"), time = '2016-01-01 2024-11-30', geo="DE")
#write_rds(trends, "filesTrends/trends_ruecken20241227.rds")
trends <- read_rds("filesTrends/trends_ruecken20241227.rds")

#transform important variables 
trends$interest_over_time  <- trends$interest_over_time |>
  mutate(date=as.Date(date))

#Visualize Backpain over time
trends$interest_over_time |>
  ggplot() +
  geom_line(aes(x=date, y=hits),  color = "darkred") +
  geom_smooth(aes(x=date, y=hits)) +
  geom_vline(xintercept=c(as.Date('2020-03-01'), as.Date('2022-04-01'))) +
  theme_minimal() +
  scale_colour_viridis_d(option="viridis") +
  labs(y="relative amount of queries", x="year", title="Figure 1: Queries about back pain over time")

Figure 1 shows that the relative number of queries steadily increased over time before the lockdown, indicating a positive time trend. During the lockdown period (marked with the 2 vertical lines) back pain-related queries increased even more – likely due to the lockdown – peaking at the end of 2022 before declining afterward. To better understand the data structure before the treatment let us decompose trend, seasonality, and error (using just complete 12-year periods):

#create time series:
ts_in <- trends$interest_over_time |>
  filter(date>= '2016-01-01' & date<= '2019-12-31') |>
  mutate(Jahr=year(date)) |>
  select(hits) |> 
  pull()

back_ts <- ts(ts_in, start = c(2016, 1),   frequency = 12) 

dcp <- mstl(back_ts, s.window="periodic")

autoplot(dcp) +
  labs(title="Figure 2: Multiple seasonal decomposition of back pain queries")

A monthly seasonal pattern can be seen which we should keep in mind when analyzing the relationship between the pandemic and back pain. Also, due to the time structure of our data, autocorrelation is a potential issue we should be aware of.

ggAcf(back_ts) +
  labs(title="Figure 3: Autocorrelation plot (back pain queries)")

A visual inspection of Figure 3 reveals a strong autocorrelation in the time series across up to 13 monthly lags. Let us now assess whether the COVID-19 period is statistically associated with Google queries for back pain. We apply segmented regression analysis (e.g. Wagner et al. 2002; Jebb et al. 2015; Xiao, Augusto, and Wagenaar 2021) assuming an interrupted linear time trend in our model, using data from 2016 until the end of March 2022. This period includes the most rigorous lockdown policies.

The dependent variable, \(hits_t\), represents the relative amount of queries for back pain over months \(t\) and is considered in logarithmic form. The explanatory integer variable \(time_t\) accounts for an assumed (partly counterfactual) linear time trend. Since lockdown policies began in Germany on 2020-3-16, we test for a level change after the lockdown’s start by adding a dummy variable \(A_t\) to the specification. To test for gradual changes of interest in back pain during the lockdown period, we include the interaction term \(A_t\times (time_t-\tau)\), \(\tau\), indicating the time index for the first lockdown. This leads to the following regression equation:

\(log(hits_t) = \beta_0 + \beta_1\,time_t + \beta_2\,A_t + \beta_3\,A_t\times (time_t- \tau) + \epsilon_t\)

Finally, dummy variables for each month are added to our specification in order to capture the seasonality of queries for back pain, as discussed above. Autocorrelated residuals are addressed by applying Newey-West standard errors with a lag length of 3. Using higher lags did not change the conclusions from this analysis.

data_impact <- trends$interest_over_time |>
  filter(date<='2022-03-31') |> # restrict data on relevant time period
  arrange(date) |>
  mutate(time=row_number()-1, # set first observation of time index to zero
         month=as.factor(month(date)), hits_l1=dplyr::lag(hits),     
         after=if_else(date>=as.Date('2020-03-16'), 1, 0),
         tau=min(ifelse(after==1, time, NA_integer_), na.rm=TRUE), # minimal time index after treatment
         time_diff=time-tau, A_TimeDiff=after*time_diff) # centering, create interaction term

lm_out <- lm(log(hits) ~ time + month  + after + A_TimeDiff, data=data_impact)

ipct_out <- round(exp(lm_out$coefficients["A_TimeDiff"])*100-100, 3)

summ(lm_out, vcov=NeweyWest(lm_out, lag=3, prewhite=TRUE, adjust=TRUE), digits = 3)
Observations 75
Dependent variable log(hits)
Type OLS linear regression
F(14,60) 27.044
R² 0.863
Adj. R² 0.831
Est. S.E. t val. p
(Intercept) 4.254 0.020 207.852 0.000
time 0.002 0.001 3.029 0.004
month2 -0.004 0.019 -0.233 0.817
month3 -0.038 0.030 -1.268 0.210
month4 -0.078 0.021 -3.794 0.000
month5 -0.095 0.030 -3.151 0.003
month6 -0.144 0.016 -8.779 0.000
month7 -0.120 0.012 -9.658 0.000
month8 -0.085 0.018 -4.681 0.000
month9 -0.054 0.023 -2.392 0.020
month10 -0.022 0.017 -1.353 0.181
month11 -0.053 0.018 -2.895 0.005
month12 -0.063 0.026 -2.430 0.018
after 0.019 0.019 1.025 0.309
A_TimeDiff 0.005 0.001 4.083 0.000
Standard errors: User-specified

The level change after treatment is positive but lacks statistical significance (p>0.05). However, compared to the baseline, we find a steady increase in relative queries from the beginning of the lockdown onward (p<0.05). From this point, queries increased approximately \((e^{\hat{\beta}_3}-1)\cdot100=\)0.546 percent each month – in addition to the assumed counterfactual time trend \(\beta_1\). These findings suggest a higher increase in back pain queries over time, rather than an immediate jump in searches after the lockdown began. In summary, our regression shows that the COVID-19 period is associated with a higher share of queries for back pain, given a linear time trend and seasonal adjustments. It seems that, although people were spared from COVID-19 infections, there were negative external effects on other health outcomes in Germany, including back pain. Fortunately, we also know from above that interest in back pain began to decline from 2023 onward, nurturing the hope that the increase in back pain was a temporary phenomenon.

Conclusion

We observed a substantial increase in the relative interest in back pain based on online queries using Google Trends data from Germany. From a policy perspective, sufficient health care supply is necessary to ensure optimal treatment of back pain and avoid chronic conditions. Employers could support ergonomic resources and health campaigns for remote workers. Self-care is another critical aspect of back pain management to develop individual exercise routines and increase physical activity. Digital services can also play a role by providing easy access to relevant information.

One thing we should keep in mind is that Google Trends reflects the relative importance of queries compared to all queries at a specific location and time. Overall internet queries increased strongly in 2020, so our results suggest back pain queries increased even more strongly. A limitation of this analysis is that higher online searches for back pain do not necessarily equate to an increase in back pain prevalence, but could indicate a higher awareness about this topic. Assuming a higher prevalence, the declining interest in back pain from 2023 onward gives hope that the increase may have been a temporary phenomenon during the pandemic.

In addition, the analysis shows the value of process-data from internet companies in answering important questions. As Stephens-Davidowitz (2018) argues, internet searches often reflect people’s thoughts and motivations more honestly than survey responses.

References

Brodeur, Abel, Andrew E Clark, Sarah Fleche, and Nattavudh Powdthavee. 2021. “COVID-19, Lockdowns and Well-Being: Evidence from Google Trends.” Journal of Public Economics 193: 104346.
Ciaffi, Jacopo, Riccardo Meliconi, Maria Paola Landini, Luana Mancarella, Veronica Brusi, Cesare Faldini, and Francesco Ursini. 2021. “Seasonality of Back Pain in Italy: An Infodemiology Study.” International Journal of Environmental Research and Public Health 18 (3): 1325.
Hoerger, Michael, Sarah Alonzi, Laura M Perry, Hallie M Voss, Sanjana Easwar, and James I Gerhart. 2020. “Impact of the COVID-19 Pandemic on Mental Health: Real-Time Surveillance Using Google Trends.” Psychological Trauma: Theory, Research, Practice, and Policy 12 (6): 567.
Hyndman, Rob, George Athanasopoulos, Christoph Bergmeir, Gabriel Caceres, Leanne Chhay, Mitchell O’Hara-Wild, Fotios Petropoulos, Slava Razbash, Earo Wang, and Farah Yasmeen. 2021. forecast: Forecasting Functions for Time Series and Linear Models. https://pkg.robjhyndman.com/forecast/.
Jebb, Andrew T, Louis Tay, Wei Wang, and Qiming Huang. 2015. “Time Series Analysis for Psychological Research: Examining and Forecasting Change.” Frontiers in Psychology 6: 727.
Knipe, Duleeka, Hannah Evans, Amanda Marchant, David Gunnell, and Ann John. 2020. “Mapping Population Mental Health Concerns Related to COVID-19 and the Consequences of Physical Distancing: A Google Trends Analysis.” Wellcome Open Research 5.
Long, Jacob A. 2020. Jtools: Analysis and Presentation of Social Scientific Data. https://cran.r-project.org/package=jtools.
Massicotte, Philippe, and Dirk Eddelbuettel. 2021. gtrendsR: Perform and Display Google Trends Queries. https://CRAN.R-project.org/package=gtrendsR.
Stephens-Davidowitz, Seth. 2018. Everybody Lies: What the Internet Can Tell Us about Who We Really Are. Bloomsbury Publishing.
Szilagyi, Istvan-Szilard, Torsten Ullrich, Kordula Lang-Illievich, Christoph Klivinyi, Gregor Alexander Schittek, Holger Simonis, and Helmar Bornemann-Cimenti. 2021. “Google Trends for Pain Search Terms in the World’s Most Populated Regions Before and After the First Recorded COVID-19 Case: Infodemiological Study.” Journal of Medical Internet Research 23 (4): e27214.
Wagner, Anita K, Stephen B Soumerai, Fang Zhang, and Dennis Ross-Degnan. 2002. “Segmented Regression Analysis of Interrupted Time Series Studies in Medication Use Research.” Journal of Clinical Pharmacy and Therapeutics 27 (4): 299–309.
Xiao, Hong, Orvalho Augusto, and Bradley H Wagenaar. 2021. “Reflection on Modern Methods: A Common Error in the Segmented Regression Parameterization of Interrupted Time-Series Analyses.” International Journal of Epidemiology 50 (3): 1011–15.
Zeileis, Achim, Susanne Köll, and Nathaniel Graham. 2020. “Various Versatile Variances: An Object-Oriented Implementation of Clustered Covariances in R.” Journal of Statistical Software 95 (1): 1–36. https://doi.org/10.18637/jss.v095.i01.