Research in Social and

Administrative Pharmacy j (2014) j–j

Original Research

Web search query volume as a measureof pharmaceutical utilization and changes

in prescribing patternsJacob E. Simmering, M.S.a, Linnea A. Polgreen, Ph.D.a,*,

Philip M. Polgreen, M.D., M.P.H.b,caDepartment of Pharmacy Practice and Science, University of Iowa, Iowa City, IA 52242, USA

bDepartment of Internal Medicine, University of Iowa, Iowa City, IA, USAcDepartment of Epidemiology, University of Iowa, Iowa City, IA, USA

Abstract

Background: Monitoring prescription drug utilization is important for both drug safety and drugmarketing purposes. However, access to utilization data is often expensive, limited and not timely.Objectives: To demonstrate and validate the use of web search engine queries as a method for timely

monitoring of drug utilization and changes in prescribing behaviors.Methods: Drug utilization time series were obtained from the Medical Expenditure Panel Survey andnormalized search volume was obtained from Google Trends. Correlation between the series was estimated

using a cross-correlation function. Changes in the search volume following knowledge events were detectedusing a cumulative sums changepoint method.Results: Search volume tracks closely with the utilization rates of several seasonal prescription drugs.

Additionally, search volume exhibits changes following known major knowledge events, such as thepublication of new information.Conclusions: Search volume provides a first order approximation to pharmaceutical utilization in thecommunity and can be used to detect changes in prescribing behavior.

� 2014 Elsevier Inc. All rights reserved.

Keywords: Pharmacovigilance; Post-marketing surveillance; Novel data sources

Introduction

Accurate and timely estimates of pharmaceu-tical utilization as well as changes in interest or

demand for pharmaceuticals are critical for manydrug safety and marketing related investiga-tions.1,2 For example, the cumulative burden of

an adverse drug event is the result of not only

* Corresponding author. Tel.: þ1 319 384 3024; fax: þ1 3

E-mail address: linnea-polgreen@uiowa.edu (L.A. Polgreen

1551-7411/$ - see front matter � 2014 Elsevier Inc. All rights

http://dx.doi.org/10.1016/j.sapharm.2014.01.003

the relative frequency of that particular eventbut also the number of people taking the drug.Knowing the expected number of people at risk

for a particular adverse event, especially giventhe potential for post-marketing novel events, isimportant for designing and guiding interventions

for drug safety.1,2 Additionally, marketing and

19 353 5646.

).

reserved.

2 Simmering et al. / Research in Social and Administrative Pharmacy j (2014) 1–8

other efforts to increase awareness of a specificdrug are heavily dependent on detecting anychanges in the utilization of the targeted drug.

Historically, data regarding the use of phar-maceuticals has been limited in geographic scope,drawn from a relatively small sample, expensive,difficult to obtain, or not widely available in a

timely fashion.3–5 For example, data from IMSHealth has no listed price and users are directedto sales staff for a quote.5 Data from more easily

accessible sources such as the Medical Expendi-ture Panel Survey is only published after consider-able delay. In addition, there is not a national

reporting system in place for prescription drugutilization2 as there is for influenza.6 As a result,other than data from individual, national phar-macy chains, there are no geographically diverse,

current and available estimates of utilization. Aconsequence of this delay is the information istoo old to lead to timely modifications to existing

interventions or marketing efforts. These datalimitations complicate research and marketingefforts.

Investigators in various fields have usedInternet search volume as a proxy for consumerinterest and have used these data to help forecast

sales of consumer goods. For example, Cho andVarian have shown a strong correlation betweensearch volume at Google and retail sales by thetype of good, brand specific sales of automobiles,

house sales and even travel.7 Search volume atYahoo has been shown to track with the numberof sales of specific music tracks and movie box of-

fice receipts.8 The methods have been extended tohealth related search topics, most notably influ-enza and other infectious diseases.9–11 A recent

paper using search query data captured by anInternet-search-engine toolbar has shown someevidence for the ability of search queries to detectpotential adverse drug events.12

However, these methods have not yet beenwidely used to explore utilization and changes inutilization of prescription medications. Prescrip-

tion medications have some unique propertiesthat differentiate them from other consumergoods. Consumers are less likely to search for

low prices on prescription drugs compared toother consumer goods and the demand for med-ications is induced by a condition the patient did

not desire.13 Consumers choose to go to a movieor buy a new car but either do not have or havemuch less agency over their prescription druguse. Additionally, access to such drugs is gated

through a prescriber and dispenser. Search

volume based methods of surveillance and estima-tion have not yet been validated on prescriptiondrugs.

Our goal is to demonstrate a relationshipbetween search volume and a nationally represen-tative estimate of actual medication use drawnfrom the Medical Expenditure Panel Survey

(MEPS). Additionally, we examined the responsesin search volume following major events expectedto change the prescribing and demand patterns

(We refer to these as “pharmaceutical knowledgeevents”) to determine if changes in search volumecorrespond to these knowledge events.

Methods

Google Trends

Search volume data were obtained from Goo-gle Trends. Google Trends is a publicly available

source that provides normalized Google “searchshare” values by week in the range 0–100 startingin January of 2004 for a set of common queries.7

The absolute number of searches represented by agiven value is not necessarily constant across time,as the total number of searches, and thus search

share, may change.7 However, assuming that thetotal number and relative frequency of searchesis roughly constant, changes in search share canbe interpreted as changes in the absolute number

of searches for a given query. A search volumeof 0 means that in the given week the providedset of keywords were not popular enough to

have been indexed. Unless otherwise noted, key-words included both the generic name andcommonly used trade names.

Validation of search volume as a measure ofcommunity utilization

In order to determine if search volume is areasonable proxy for drug utilization, we first

extracted weekly outpatient drug-utilization esti-mates from the Medical Expenditure Panel Survey(MEPS) for 2004–2009 for nine drugs fromvarious therapeutic classes. MEPS is a yearly,

nationally representative panel survey of thecommunity based, non-institutionalized popula-tion and includes information about prescription

drug use.14 The drugs were chosen for an a prioriexpectation of seasonal behavior. Some drugs areused seasonally which creates within-year varia-

tion that allows for meaningful comparison ofactual utilization and search volume estimates.We selected amoxicillin, azelastine, azithromycin,

3Simmering et al. / Research in Social and Administrative Pharmacy j (2014) 1–8

benzonatate, cefdinir, ciprofloxacin, levofloxacin,moxifloxacin and olopatadine. All of these drugsare used to treat infections or allergies with aknown seasonal pattern and, as such, the drugs

are expected to exhibit highly seasonal time series.The included drugs include examples expected topeak either during the summer or the winter.

Dispensing events for each of the nine drugslisted above were found via regular expressionsmatching of the MEPS pharmacy reported name

and the generic or trade names of the targetdrugs. Regular expressions are a formalized andflexible method for string and substring match-

ing.15 Prescription events were found by match-ing on key substrings of the drug name (e.g., aset of 4–6 characters) to abstract the set of namesused for that drug in the MEPS pharmacy-

reported-name variable. The resulting list was re-viewed to ensure against false positive matches(e.g., levothyroxine and levofloxacin both match-

ing the substring “levo”). After removing anyfalse positives, the list of names was then usedas a “gold standard” and all records with match-

ing pharmacy reported names were extractedfrom the MEPS drug dataset. The search volumewas obtained over the same interval using Goo-

gle Trends. The cross-correlation function(CCF) between the MEPS-derived medicationutilization and the Google search volume wascalculated. The CCF is a series of correlations

calculated between two series with various lagsin one of the series. For example, at a lag ofzero, the CCF is the common correlation be-

tween the two series. At lag of þ1, the CCF isthe correlation between the two series with thesecond series shifted by one unit of time (e.g.,

the correlation between a given week in the firstseries and the following week in the other series).

Table 1

Drugs potentially subject to major changes in prescribing beh

Affected drug Alternate drug

Omacor Lovaza

Kapidex Dexilant

Zicam Cold Remedy Nasal Products N/A

Rosiglitazone (Avandia) Pioglitazone (

COX-2 Selective Inhibitors N/A

Vytorin/ezetimibe (Zetia)a Simvastatin (Z

Pseudoephedrine Phenylephrine

a Simvastatin, the generic name for part of the Vytorin co

series as the goal was to measure use of ezetimibe alone or in co

have captured use of simvastatin without ezetimibe or combin

Knowledge events

To determine if “pharmaceutical knowledgeevents” can be detected in search data, we used adifferent set of pharmaceutical agents than we

used in our validation described above. Ourpharmacist reviewer reviewed the list of FDAsafety alerts and press releases from third parties(e.g., Heart.org) and identified knowledge events

that were likely to have resulted in a major changein practice. These sources were supplemented withthe pharmacist’s knowledge of changes in the field

that may not have been captured by either of theinputs. In the end, our pharmacist identified sevenmedications that were subject to some major

knowledge event between 2004 and 2011. Reasonsfor changes in prescribing behavior included namechanges, recalls, major safety alerts or efficacy in-formation. We included the two drugs where the

name of the medication was changed by theFDA after it was released. These drugs providea positive control for our methods in that a

change in behavior must occur as the drug is nolonger ordered or dispensed under the old name.Additionally, as the name changes were essentially

instant, they allow us to estimate the minimumamount of time for a change to occur withoutstanding supplies. The included drugs, event

dates and reason for the event are listed inTable 1. We believe this list provides a reasonablesample of the different possible causes of knowl-edge events and includes positive controls making

it a reasonable test set for determining if knowl-edge events are detectable in search volume.

The Google Trends search volume was

abstracted for each series and changepoints werelocated. A changepoint is a point at which thetime series’ behavior changes in a major way. For

avior

Date of event Event type

08/01/2007 Name change

03/04/2010 Name change

06/16/2009 Safety alert

Actos) 05/21/2007 Safety alert

09/30/2004 Safety alert

ocor) 01/14/2008 Efficacy

03/09/2006 Regulatory

mbination, was not included in the search string for this

mbination with simvastatin. Including simvastatin would

ation drugs (e.g., Vytorin).

4 Simmering et al. / Research in Social and Administrative Pharmacy j (2014) 1–8

example, a series that increased steadily for anumber of months and then stopped increasinghas a changepoint where the series plateaus. The

changepoints for each series were located using acumulative sums method. Formally, for each weekt with search volume vt in the series, the cumula-tive sum is St ¼

Pti¼1vi � v. If the series follows

a constant process over the interval under study,the observations vi and viþk where k s 0 are inde-pendent. If the observations are independent, we

would expect the deviations to “cancel” eachother out and EðStÞ ¼ 0. The week t that maxi-mizes the absolute value of the cumulative sum

is taken as the most likely changepoint locationtc in the interval [1,T] where T is the end of the se-ries. As this method will always return the mostprobable changepoint in the interval [1,T], we

bootstrapped an empirical P-value for the result-ing changepoint. If tc was a significant change-point at a ¼ 0.05, we partitioned the series into

the intervals [1,tc] and [tc,T] and repeated this pro-cess recursively until no new significant change-points were found.

Once the probable changepoint nearest theknowledge event was found, the rate at whichthe search volume for the affected drug was

replaced by search volume for the alternate drugwas estimated. This is known as the marginal rateof substitution (MRS). The MRS was estimatedby regressing the affected drug’s search volume on

the volume for the alternative drug over the 12months before and after the changepoint. Simplelinear regression failed to produce meaningful

estimates due to the relatively small number ofpoints and the relatively high number of outliers.By definition, the volume for weeks near the

changepoint is very volatile. To minimize theeffect of this volatility on the regression model,iteratively re-weighted least squares was used toreduce the influence of the outliers and produce

more accurate estimates of the MRS. Iterativelyre-weighted least squares is a form of regressionwhere the regression weights are selected via an

iterative process to minimize the effect of anoutliers and better meet the assumptions of con-stant variance.

Results

Utilization

Of the nine seasonal drugs we considered todetermine if search volume is a reasonable mea-sure of drug utilization, only three (amoxicillin,

azithromycin and cefdinir) had enough outpatientdispensing events in the MEPS data to construct atime series suitable for analysis. The other 6 drugs

(azelastine, benzonatate, ciprofloxacin, levofloxa-cin, moxifloxacin and olopatadine) had manyweeks with 0 observed fills. This is due to therelatively low rate of use of these drugs, especially

compared to the three more-common series, com-bined with the moderate sample size of MEPS(roughly 30,000 per year). Additionally, these

series had very high levels of inter-week variancethat dwarfed the expected seasonal variance.When the MEPS sample weights are applied to

produce national level estimates, the week-to-week variance explodes. Because of these limita-tions with azelastine, benzonatate, ciprofloxacin,levofloxacin, moxifloxacin and olopatadine, we

only used amoxicillin, azithromycin and cefdinirin our cross-correlation analysis.

The cross-correlation functions for the three

considered drugs showed positive correlationbetween the search volume and MEPS-derivedoutpatient utilization rates at lags near 0 (Fig. 1).

In other words, significant correlation existsbetween the MEPS-derived use and the searchvolume in the same week. There is also a

strong positive relationship at year intervalsand a strong negative relationship at half-yearintervals.

Knowledge events

Each of the seven series that we used toexamine knowledge events showed a significantchangepoint in the search series that coincided

with the knowledge-event date. For example, inJanuary 2008, it was reported that combinationVytorin/ezetimibe did not outperform treatment

with just simvastatin. Following this event, asignificant change in the Vytorin series occurred,Fig. 2. Vytorin/Ezetimibe lost search share rela-

tive to simvastatin immediately after a spike insearch interest in January 2008.

A similar pattern was observed in the otherseries corresponding to the different knowledge

events. Typically a spike in search volume wasassociated with the timing of the “knowledgeevent” followed by a changepoint in the affected

series. The changepoints were all near the eventdate in time, Table 2. For the drugs with an ex-pected alternative drug, there was an increase

the searches for the alternative drug followingthe event date. This would suggest a change inconsumer and provider interest in a drug

Fig. 1. Cross-correlation functions for amoxicillin, azithromycin and cefdinir utilization and Google Trends volume.

The height of the bar indicates the correlation coefficient between the MEPS-derived volume series and the shifted Goo-

gle Trends series. Horizontal lines denote a ¼ 0.05 significance levels, anything outside of the lines is significantly

different from 0. Significant positive correlations are found near 0 and �1 year. Negative correlations are found at

�0.5 and 1.5 years.

5Simmering et al. / Research in Social and Administrative Pharmacy j (2014) 1–8

following the “knowledge event.” The rate ofchange and interval was similar between knowl-

edge event types.Finally, the marginal rate of substitution be-

tween the primary affected drug and a likelyalternate is also shown in Table 2. The estimates

for nearly all of the drugs suggest a decrease insearch volume for the affected drug followingthe changepoint. There is a related increase in

search volume for the probable alternate drug inall but one case. Rosiglitazone/pioglitazone areoutliers in this respect as the marginal rate of sub-

stitution is positive.

Discussion

These results show that search volume fordrug-related keywords provides an accurate and

timely first-order approximation of actual com-munity utilization. Moreover, search volumechanges occurred after major knowledge events

with minimal delay. The ability to both estimatethe utilization at a given time, including historicalutilization, and the sensitivity of drug searches to

major shifts in interest suggest many future po-tential applications of this methodology for drugmarketing and pharmacovigilance efforts.

Significant, strong correlation at zero lag be-tween the MEPS-derived utilization and the

Google Trends search share was found for thethree seasonal drugs considered (amoxicillin, azi-thromycin and cefdinir). Also, each CCF had alocal minima at �6 months and local maxima at

�1 year. As these drugs are strongly seasonal, astrongly negative correlation would be expected atthe half-year mark and the same week in previous

and subsequent years would be expected to bepositively correlated. This lends support to inter-pretation that the two series measure the same

phenomena.Internet searches were also correlated with

knowledge events. For example, following thepublication of new efficacy information in

January 2008, the search volume for Vytorin/ezetimibe started to decline. The search volumefor simvastatin, a likely alternative, increased

following the event date. This pattern of “swap-ping” for the expected alternative drug wasobserved for nearly all of the considered drug

pairs.Notably, for the two drugs that underwent

relabeling, changes in Internet searches appeared

roughly 2 months after relabeling. This interval islikely the byproduct of 30- and 90-day suppliesbeing standard fills for many chronic drugs. It

Fig. 2. Log Google Trends search volume for Vytorin/ezetimibe/Zetia and simvastatin/Zocor. The black vertical line

marks the knowledge event of Jan 14, 2008. Following the release of the new efficacy information in Jan 2008, we see

an increase in interest in Vytorin/ezetimibe/Zetia followed by a rapid decrease to a new baseline lower than that previ-

ously observed. The simvastatin/Zocor series shows a slight increase at the publication date and then moves to a new

baseline and does not decrease as the Vytorin/ezetimibe/Zetia series does reflecting a possible substitution in interest be-

tween the two series following the Jan 2008 event.

6 Simmering et al. / Research in Social and Administrative Pharmacy j (2014) 1–8

would take roughly 1–2 months before enoughpatients have had their medications relabeled orchanged. In addition, this method detected theswitch away from COX-2 inhibitors and Vytorin

in approximately 3 months, suggesting relativequick detection of these knowledge events.

This pattern of switching after 2 months did

not occur for rosiglitazone and pioglitazone. Inthe interval near the changepoint, the MRS wasgreater than zero. However, this drug pair was

much more sensitive to the specification of theweights in the regression than the other combina-tions. However, it is not clear that pioglitazone is

Table 2

Event dates and CUSUM changepoints

Series Date of event Chang

Omacor 08/01/2007 09/23/

Kapidex 03/04/2010 05/09/

Zicam Cold Remedy Nasal Products 06/16/2009 10/11/

Rosiglitazone 05/21/2007 11/11/

COX-2 Selective Inhibitors 09/30/2004 12/26/

Vytorin/ezetimibe (Zetia) 01/14/2008 04/13/

Pseudoephedrine 03/09/2006 08/27/

the best alternative drug. Pioglitazone has beenthe subject of several safety alerts and there werequestions about its safety at the time of therosiglitazone alerts.16 Given the unclear safety

profile of pioglitazone, providers may have beenreluctant to move patients from rosiglitazone topioglitazone. Providers may have decided, given

the lack of an alternative drug, to discontinuetherapy or replace rosiglitazone with anotherdrug other than pioglitazone. The other consid-

ered drug pairs did not have the problem wherethe alternative drug is also subject to major safetyquestions.

epoint Interval (days) Marginal rate of substitution

2007 52 �0.58 (�0.64, �0.52)

2010 66 �0.55 (�0.62, �0.48)

2009 116 N/A

2007 173 1.43 (1.19, 1.68)

2004 87 N/A

2008 91 �1.15 (�1.47, �0.83)

2006 170 �1.18 (�1.24, �1.13)

7Simmering et al. / Research in Social and Administrative Pharmacy j (2014) 1–8

There are drawbacks to using search volume tocharacterize medication utilization. The elderlyare the largest consumers of medications and alsounderrepresented among users of search engines

resulting in a potential mismatch between theusers of the medications and those generating thesearch data.17,18 However, we theorize that the

elderly may have younger caregivers and familymembers who turn to Google for drug informa-tion. Additionally, there are a number of limita-

tions imposed by the nature of the data fromGoogle Trends. The tool reports a normalizedshare making it possible for the same number of

total searches at different times to have twodifferent volume estimates. These properties ofGoogle Trends data make conversion to an abso-lute scale difficult, if not impossible.

Finally, the correlation between Google searchvolume and actual use as defined in our studydepends on the MEPS data. The MEPS sample is

nationally representative; however, many drugseries have high inter-week variance. The numberof observed fills in MEPS may only vary by 1 or 2

fills per week but when the weights are applied thisbecomes differences on the order of 10,000 fills perweek. This can make it difficult to construct

meaningful series for even relatively commondrugs given the approximately 30,000 subjectsample size of MEPS. Therefore, we are limitedin the number of drugs for which we can estimate

“ground truth” utilization using MEPS. Addi-tionally, the MEPS drug use time series werederived from pharmacy reported names on fills. It

is possible that a pharmacy used a very obscurename that contained no elements of the genericname or typical brand names. Patients may also

fill, but not take, a prescription.Future applications of these methods should

include an exploration of response in searchvolume following FDA safety alerts such as Black

Box Warnings and Risk Evaluation and Mitiga-tion Strategies (REMS). For example, do patientssearch for drug information when receiving a drug

with an REMS notification? Does interest spikewith a new Black Box Warning? Is this increasedawareness or interest maintained? What effect is

there on prescribing following the release of aBlack Box Warning or REMS?

Conclusions

In spite of the limitations, it appears thatGoogle Trends does provide a reasonable

characterization of population medication utiliza-tion in near real-time without requiring expensiveand time-consuming investigation. Additionally,changepoint analysis using Google Trends search

volume appears to be a timely method for thedetection of knowledge events – major shifts innot only medication interest but also utilization.

The real-time measure of interest and demand fora given medication provides a unique insight intoconsumer and provider behavior. Thus, it is likely

that the value of real-time surveillance over adiverse population at a limited cost exceeds thedrawbacks of this approach.

Acknowledgments

Ben Urick, Pharm.D. provided the list of

major knowledge events used in this study.

References

1. Aborn J. Evaluating drug effects in the post-Vioxx

World: there must be a better way. Circulation

2006;113:2173–2176.

2. Wysowski DK, Swartz L. Adverse drug event surveil-

lance the drug withdrawals in the United States.Arch

Intern Med 2005;165:1363–1369.

3. Patwardhan A, Bilkovski R. Comparison: flu pre-

scription sales data from a retail pharmacy in the

US with Google Flu Trends and US ILINet (CDC)

Data as Flu Activity Indicator. PLoS One 2012;7:8.

4. AitkenM, Berndt ER, Cutler DM. Prescription drug

spending trends in the United States: looking beyond

the turning point.Health Aff 2008;28(1):w151–w160.

5. Global Measures of Dispensing Activity. IMS

Health. http://www.imshealth.com/portal/site/ims

health/menuitem.3e17c48750a3d98f53c753c71ad8c

22a/?vgnextoid¼c5f6e590cb4dc310VgnVCM100000

a48d2ca2RCRD&vgnextfmt¼default; Accessed

01.05.13.

6. Centers for Disease Control and Prevention. Over-

view of Influenza Surveillance in the United States,

Updated October 2013. http://www.cdc.gov/flu/

weekly/overview.htm. Accessed 01.05.13.

7. Choi H, Varian H. Predicting the Present with Google

Trends, Published April 2009. http://static.googleu

sercontent.com/external_content/untrusted_dlcp/www.

google.com/en/us/googleblogs/pdfs/google_predict

ing_the_present.pdf. Accessed 01.05.13.

8. Goel S, Hofman JM, Lahaie S, Pennock DM,

Watts DJ. Predicting consumer behavior with web

search. Proc Natl Acad Sci U S A 2010;107:17486–

17490.

9. Polgreen PM, Chen Y, Pennock DM, Nelson FB,

Weinstein RA. Using internet searches for influenza

surveillance. Clin Infect Dis 2008;47:1443–1448.

8 Simmering et al. / Research in Social and Administrative Pharmacy j (2014) 1–8

10. Ginsberg J, Mohebbi MH, Patel RS, Brammer L,

Smolinski MS, Brilliant L. Detecting influenza epi-

demics using search engine query data. Nature

2008;457:1012–1014.

11. Chan EH, Sahai V, Conrad C, Brownstein JS. Using

web search query data to monitor dengue epidemics.

PLoS Negl Trop Dis 2011;5:e1206.

12. White RW, Tatonettie NP, Shah NH, Altman RB,

Horvitz E. Web-scale pharmacovigilance: listening

to signals from the crowd. J Am Inform Med Assoc

2013;20:404–408.

13. Sorensen AT. An empirical model of heterogeneous

consumer search for retail prescription drugs. NBER

Working Paper Series; 2001.

14. Agency for Healthcare Research and Quality. Survey

Background, Updated August 2009. http://meps.

ahrq.gov/mepsweb/data_stats/data_overview.jsp. Ac-

cessed 01.05.13.

15. R Core Team. Regular Expressions as Used in R,

Updated April 2013. http://stat.ethz.ch/R-manual/

R-devel/library/base/html/regex.html. Accessed

December 2013.

16. US Food and Drug Administration. Actos (pioglita-

zone): Ongoing Safety Review – Potential Increased

Risk of Bladder Cancer, Updated August 2011.

http://www.fda.gov/Safety/MedWatch/SafetyInfor

mation/SafetyAlertsforHumanMedicalProducts/ucm

226257.htm. Accessed 01.05.13.

17. ZickuhrK,MaddenM.Older Adults and the Internet.

Pew Internet and American Life Project; Published

June 2012. http://www.pewinternet.org/w/media//

Files/Reports/2012/PIP_Older_adults_and_internet_

use.pdf, Accessed 01.05.13.

18. Gu Q, Dillon CF, Burt VL. Prescription drug use

continues to increase. NCHS Data Brief 2010;42:

1–8.