Skip to main content
NCBI home page
NIHPA Author Manuscripts logoLink to NIHPA Author Manuscripts
. Author manuscript; available in PMC: 2025 Oct 1.
Published in final edited form as: J Psychiatr Res. 2024 Aug 3;178:219–224. doi: 10.1016/j.jpsychires.2024.08.001

A Large-Scale Observational Comparison of Antidepressants and their Effects

Michael V Heinz 1,5,6, Elad Yom-Tov 3,4, Daniel M Mackin 1,5,6, Rina Matsumura 1, Nicholas C Jacobson 1,2,5,6
PMCID: PMC11398883  NIHMSID: NIHMS2017855  PMID: 39163659

Abstract

Background

Selective Serotonin Reuptake Inhibitors (SSRIs) represent a diverse class of medications widely prescribed for depression and anxiety. Despite their common use, there is an absence of large-scale, real-world evidence capturing the heterogeneity in their effects on individuals. This study addresses this gap by utilizing naturalistic search data to explore the varied impact of six different SSRIs on user behavior.

Methods

The study sample included ~508 thousand Bing users with searches for one of six SSRIs (citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, sertraline) from April-December 2022, comprising 510 million queries. Cox proportional hazard models were employed to examine 30 topics (e.g., shopping, tourism, health) and 195 health symptoms (e.g., anxiety, weight gain, impotence), using each SSRI as a reference. We assessed the relative hazard ratios between drugs and, where feasible, ranked the SSRIs based on their observed effects. We used Cox proportional hazard models in order to account for both the likelihood of users searching for a particular topic or symptom and the associated time to that search. The temporal aspect aided in distinguishing between potential symptoms of the disorder, short-term medication side effects, and later appearing side effects.

Results

Differences were found in search behaviors associated with each SSRI. E.g., fluvoxamine was associated with a significantly higher likelihood of searching weight gain compared to all other SSRIs (HRs 1.85–2.93). Searches following citalopram were associated with significantly higher rates of later impotence queries compared to all other SSRIs (HRs 5.11–7.76), except fluvoxamine. Fluvoxamine was associated with a significantly higher rate of health related searches than all other SSRIs (HRs 2.11–2.36).

Conclusions

Our study reveals new insights into the varying SSRI impacts, suggesting distinct symptom profiles. This novel use of large-scale, naturalistic search data contributes to pharmacovigilance efforts, enhancing our understanding of intra-class variation among SSRIs, potentially uncovering previously unidentified drug effects.

Introduction

Selective serotonin reuptake inhibitors (SSRIs) are a common, and often effective, first-line treatment for depressive, anxiety, and related disorders (Boschloo et al., 2023; Jakobsen et al., 2017; Jakubovski et al., 2019; Murphy et al., 2021). Despite a common proposed mechanism of action (involving the 5-hydroxytryptamine [5-HT] neuronal reuptake transporter), SSRI medication classes differ in their secondary binding properties. Specifically, classes of SSRIs vary in their action at noradrenergic, dopaminergic, muscarinic cholinergic, histaminergic, and sigma receptors, as well as in their inhibition of the cytochrome P450 enzymes (Stahl, 2004). These differences in secondary properties may be associated with differential treatment responses, as well as differences in the side effects (Stahl, 2004). However, much of what is known about differential treatment response to, and side effects of, SSRIs comes from small or moderate placebo-controlled trials, often using self-reported data; for instance, 715 and 429 patients taking escitalopram with MDD and GAD, respectively, were included in adverse event analysis in the package insert (“Lexapro Package Insert,” 2017). While these sample sizes adequately detect positive effects of the drug, they are underpowered to detect side effects (Ferguson, 2001; Rief et al., 2011) or subtle differences between drugs. Importantly, SSRIs are often prescribed using a trial and error approach (Taliaz et al., 2021), partly due to the lack of information about side effects and naturalistic correlates. This approach to pharmacological intervention is inefficient and likely less effective than treatment matching (Taliaz et al., 2021), contributing to increased attrition and noncompliance with medications.

Over the last decade, it has become feasible to examine large quantities of naturalistic data collected from internet-based sources. These unstructured data, such as Bing search queries, provide a rich source of explorable information shown to be representative of the larger population (Yom-Tov and Gabrilovich, 2013). Critically, these data may provide insight into the personalized real-world impact of psychotropic medications. Many studies to date have used social media platforms, such as Twitter and Facebook, for pharmacovigilance (Chokor et al., 2016; Coupland et al., 2011; Pappa and Stergioulas, 2019; Saha et al., 2021, 2019), defined as the identification, assessment, and response to drug-related safety issues (Gelfand and Noe, 2021). For example, Saha, et. al. (2019, 2021) used large Twitter datasets, comprising 8 million-300 million posts, to explore the distinctive effects of psychiatric drugs using language markers. A similar study (Song and Xie, 2020) using Reddit supported the use of social media for characterizing the effects of specific psychiatric medications, including antidepressants. With unprecedented sample sizes, internet-powered studies have the potential to detect subtle medication effects and further have the advantage of using naturalistic patient-generated data. However, these studies have not focused on direct and quantitative within-class comparison of antidepressants. Further, such studies, while large in overall sample size, are likely to lack the within-person temporal resolution for a self-controlled approach, which would optimally account for confounding variables. Problematically, there is evidence that those who do post frequently on social media (leading to higher temporal resolution) may not be representative of the general population (Panek et al., 2018), and it stands to reason that such data may be further biased given the public readership of social media posts.

User queries from internet search engines such as Bing and Google are important sources of naturalistic person-generated, and large-volume, internet data, which address many of the aforementioned limitations and provide a promising approach toward within-class antidepressant comparison. Internet search data has the advantage of greater temporal definition and generalizability given the relative expectation of privacy compared to social media posting. Additionally, such data have shown promise in providing naturalistic pharmacovigilance for adverse drug reactions (ADRs) despite requiring far fewer resources than adequately powered randomized clinical trials (Yom-Tov and Gabrilovich, 2013). This provides an affordable and effective method for examining side effects and correlates of medication use in a previously unattainable sample regarding size and naturalistic setting. Moreover, with the increasing development of many new powerful psychotropic medications, RCT evidence alone is insufficient, with the emerging field of pharmacoepidemiology (Garbe et al., 2019) emphasizing the importance of postmarket drug surveillance among large populations.

Preliminary research in this area has successfully used internet search data for pharmacovigilance to explore ADRs related to common U.S. drugs, including psychiatric drugs (Yom-Tov and Gabrilovich, 2013). That work provided evidence for the strong relationship between searches for drugs and known adverse events, supporting the validity of this approach. Search data has also demonstrated value in early detection of ADRs (White et al., 2016) and discovery of previously unknown adverse interactions involving psychiatric drugs (White et al., 2013). White et al. (2013), for example, identified hyperglycemia as an adverse interaction between paroxetine (an SSRI) and pravastatin (a cholesterol-lowering drug), which was later confirmed by the FDA.

Given the promise of search engine data for pharmaceutical monitoring, combined with the need for empirically-grounded and nuanced antidepressant profiles, we use search engine queries to explore within-class differences among SSRIs. In this context, the present work aims to explore differences between SSRIs using a large sample of over 500 million search engine queries involving one of six antidepressants. To accomplish this aim, we use first-time searches for single SSRIs as proxy markers for drug prescription and searches for health-related keywords as reflective of a user’s experience of real-world symptoms. Supporting this approach, prior work has (1) demonstrated search query data is strongly associated with antidepressant prescription behavior (Gahr et al., 2015) and (2) established a precedent for using search queries as proxies for health symptoms (Jacobson et al., 2022; Yom-Tov and Gabrilovich, 2013). After identifying persons who search for an SSRI, we then calculate the relative odds of an increase in a particular queried keyword. Given the sample size, this research is powered to detect even subtle associations between SSRI use and symptoms and behaviors, as well as differences between SSRIs. Critically, we aim to explore differences between SSRIs in an ecologically valid way, with the expectation of finding both known and previously unknown medication-specific associations.

Methods

Participants

We extracted data from all users within the United States to the Microsoft Bing search engine during 2022. Users in the study were those who searched for one (and only one) of the six SSRIs (citalopram, escitalopram, fluoxetine, fluvoxamine, paroxetine, or sertraline) during the period of April to December 2022 but did not do so in the first 3 months of 2022. The latter was done to identify users who likely asked about an SSRI for the first time.

Search Queries

We extracted all queries of the selected users for the entirety of 2022. For each query we extracted an anonymous user identifier, its time and date, and the query text. Bing users have been shown to be a representative sample of internet users (Rosenblum and Yom-Tov, 2017) with a sizeable PC market share of 39% in the United States (Microsoft, 2024). The SSRIs were extracted according to both their generic and brand names. Each query was labeled into one or more of 30 topics (e.g., shopping, tourism, and health) using a proprietary classifier developed by Bing.

Additionally, we used a keyword matching approach to identify queries containing relevant physical and psychiatric symptoms. This approach, which was developed and validated by Yom-Tov and Gabrilovich (2013) using behavioral cues in Yahoo web search, employed a medical symptom dictionary of 195 face-valid physical and psychiatric symptoms with respective synonyms. User search queries were programmatically parsed for inclusion of medical dictionary entries. A positive result occurred if either the dictionary key or associated synonyms were present. For example, “rapid heart”, “rapid heart rate”, and “fast heart rate” would constitute tachycardia. The time of each query was calculated relative to the time of the first query made by the user which mentioned an SSRI.

Main Analyses

Cox Proportional Hazard Modeling

A Cox proportional hazard model was applied to the data at a user level for each of the topics and symptoms; that is, 195+30=225 models. The independent variable of the model was the specific SSRI queried by the user. The dependent variable was whether the user asked about the topic or symptom of that model. If the user asked about the symptom or topic, the time of the first query for the dependent attribute (relative to the first SSRI query) was used. If the user did not ask about it, the observation was censored. The independent variable, namely, the specific SSRI, was represented using indicator variables. The model for each topic or symptom was trained separately with each of the SSRIs serving as the reference variable. Inline with our aim for within class comparison, this approach allowed for each SSRI to serve as a reference (control) for the others. Therefore, a total of 225*6=2508 models were trained.

The outcome of the models for each of the topics and symptoms is a table of the relative hazard of each drug to each other drug, with associated p-values. To ensure robustness, we adjusted for multiple comparisons using Bonferroni correction (Supplementary Reference 1A and Supplementary Reference 1B). We tested the assumptions underlying the proportional hazard models and found that in none of the symptom or topic models the assumptions of the models were violated (e.g., no time varying coefficients were found with a threshold of P<0.05 with Bonferroni correction). Note that due to the long run times of these tests they were conducted on a random sample of 100,000 points for each symptoms\topic.

Drug Ordering with the Kemeny-Young Method

To order the drugs by their effect, we used the Kemeny-Young method (Kenyon-Mathieu and Schudy, 2007) applied to the hazard rates table described above. If the hazard of a pair was not statistically significant it was not considered. Since more than one optimal Kemeny-Young ordering is possible, if there were fewer than 25 orderings, we manually analyzed the possible pairs and marked those orders that were equivalent.

Ethics Approval and Oversight

The data collection and analysis described herein was approved by the Behavioral Sciences Research Ethics Committee of the Technion, approval number 2018–032.

Results

Characteristics of the Study Sample

The study sample included approximately 508,000 Bing users with searches during 2022. Given the anonymous nature of these data, further baseline demographic data were unavailable. The total number of queries was approximately 510 million, with an average number of 1003 queries per user.

Overview of Primary Findings

Among health symptom search terms, after correcting for multiple comparisons, we found statistically significant differences between SSRIs in searches for anorexia, anxiety, depression, diarrhea, headache, impotence, fatigue, and weight gain. Similarly, among topic search terms, we found statistically significant differences between SSRIs in searches for celebrities, commerce, health, “how to”, image, recipes, and tech. All hazard ratios for all symptoms and topics are located in Supplementary Reference 1A and Supplementary Reference 1B, respectively. The results we highlight in the main manuscript are limited to statistically significant comparisons.

SSRI Comparisons Across Symptom Search Terms

Symptoms related to Mental Health

Anxiety

Escitalopram, fluvoxamine, and paroxetine were associated with statistically significantly higher odds of searching anxiety compared to fluoxetine (HRs 1.16–1.71), citalopram (HRs 1.34–1.97), and sertraline (HRs 1.46–2.16) (Table 1 and Supplementary Reference 2). Fluoxetine was associated with statistically significantly higher odds of searching for anxiety compared to sertraline (HR=1.26) and citalopram (HR=1.15) (Supplementary Reference 2).

Table 1.

Rankings of Drugs According to Association with Symptoms and Topics

Symptoms Topics
R Anxiety R Impotence R Weight gain R Commerce R Health
1 Fluvoxamine 1 Citalopram 1 Fluvoxamine 1 Fluvoxamine 1 Fluvoxamine
Paroxetine 2 Escitalopram 2 Citalopram Sertraline 2 Paroxetine
Escitalopram Sertraline Paroxetine Fluoxetine Escitalopram
2 Fluoxetine Paroxetine Fluoxetine 2 Citalopram 3 Fluoxetine
Citalopram Fluoxetine 3 Sertraline 3 Escitalopram Sertraline
Sertraline U * Fluvoxamine Escitalopram Paroxetine Citalopram
Open in a new tab

Note. Displays drug rankings according to their relative impact on symptom and topic queries in descending order. Rank (R) is shown in the immediate left column to each symptom or topic. Note that spatial top to bottom ordering within ranks is arbitrary. Only drugs with distinct ranks have statistically significant differences.

*

U represents unranked, meaning this drug could be in any order with regard to the others.

Depression

Fluvoxamine and paroxetine were associated with statistically significantly higher subsequent searches for depression, compared to fluoxetine (HRs 1.75–2.45) and sertraline (HRs 2.17–2.41) (Supplementary Reference 2).

Symptoms related to Physical Health and Constitution

GI and Headache

Citalopram and fluoxetine were associated with statistically significantly higher odds of searching for diarrhea compared to sertraline (HRs 1.85–2.17). Similarly, fluoxetine was associated with statistically significantly higher odds of searching for headache compared to sertraline (HR=1.57) (Supplementary Reference 2).

Impotence

Citalopram was associated with statistically significantly higher odds of later searches for impotence compared to all other SSRIs (HRs 5.11–7.76), except fluvoxamine. The relationship between fluvoxamine and citalopram was not statistically significant for impotence. (Table 1 and Supplementary Reference 2).

Fatigue

Paroxetine was associated with statistically significantly higher odds of searching tired compared to escitalopram (HR=2.28), fluoxetine (HR=2.15), and sertraline (HR=2.22). Further, citalopram was associated with statistically higher odds compared to escitalopram (HR=1.97) and sertraline (HR=1.91). Last, fluvoxamine was associated with statistically significantly higher odds of searching tired compared to escitalopram (HR=5.20) (Supplementary Reference 2).

Weight Gain

Fluvoxamine was associated with statistically significantly higher odds of searching weight gain compared to all other SSRIs (HRs 1.85–2.93). Further, both citalopram and fluoxetine were associated with statistically significantly higher odds of searching weight gain compared to escitalopram (HRs 1.31–1.58) and sertraline (HRs 1.19–1.44). Last, paroxetine was associated with statistically significantly higher odds of searching weight gain compared to escitalopram (HR=1.34) (Supplementary Reference 2).

SSRI Comparisons Across Topic Search Terms

Commerce

Both fluoxetine and sertraline were associated with statistically significantly higher odds of commerce related searches compared to paroxetine (HRs 2.23–2.33), escitalopram (HRs 2.10–2.19), and citalopram (HRs 1.14–1.19). Further fluvoxamine was associated with statistically significantly higher odds compared to paroxetine (HR=4.37), citalopram (HR=2.23), and fluoxetine (HR=1.96). Last citalopram was associated with statistically significantly higher odds compared to paroxetine (HR=1.96) and escitalopram (HR=1.85) (Supplementary Reference 2).

Health

Fluvoxamine was associated with statistically significantly higher odds of health related searches compared to all other SSRIs (HRs 2.11–2.36). Escitalopram was associated with statistically significantly higher odds compared to citalopram (HR=1.11) and sertraline (1.08). Paroxetine was associated with statistically significantly higher odds compared to sertraline (1.09) and fluoxetine (1.09) (Supplementary Reference 2).

Image Search Types

Paroxetine showed dramatically and statistically significantly higher odds of later image-related searches compared to citalopram (HR=25.68), escitalopram (HR=18.69), fluoxetine (HR=13.53) and sertraline (HR=10.80). Further, fluoxetine had statistically significantly higher odds compared to citalopram (HR=1.90), and sertraline had statistically significantly higher odds compared to both citalopram (HR=2.38) and escitalopram (HR=1.73) (Supplementary Reference 2).

Other Search Types

The odds ratios for statistically significant pairs of drugs for search types books, celebrities, how-to, and recipes are shown in Supplementary Reference 2.

Discussion

SSRIs are among the most commonly prescribed medications for psychiatric disorders, with rates of prescription more than doubling in high-income countries since the turn of the century (Hengartner, 2022; Lewer et al., 2015; OECD, 2021; Pratt et al., 2017). However, little is known about the real-world impacts of SSRIs both on symptoms and regarding side effects, as studies have mostly focused on clinical trial results with modest sample sizes. Data gathered naturalistically through the internet, including through search engines like Bing, are capable of monitoring the utility, side effects, and behaviors associated with these interventions (Chokor et al., 2016; Coupland et al., 2011; Gahr et al., 2015; Pappa and Stergioulas, 2019; Saha et al., 2021, 2019; Song and Xie, 2020; Yom-Tov and Gabrilovich, 2013). Therefore, the present investigation examined Bing search data from more than a half-million users who made over 500 million search queries during 2022 to identify relationships between searching for specific SSRIs and medical and mental health symptoms/side effects, and other search behaviors. Our results indicated statistically significant associations between searches for specific SSRIs and searches for mental health symptoms, including depression and anxiety, and physical health symptoms, including headaches, impotence, feeling tired, and weight gain. In many cases, the symptoms associated with specific SSRIs were consistent with prior research identifying those as side effects of the medication (e.g., consistent with existing literature, paroxetine was worse for fatigue than other SSRIs; not evident in existing literature, sertraline showed lower searches for gastrointestinal side effects). Several of the significant associations were especially intriguing and warrant further discussion.

Regarding mental health symptom queries, paroxetine and fluvoxamine showed statistically significantly higher odds of later searches for anxiety compared to all other SSRIs; among remaining SSRIs, fluoxetine showed increased odds of searching for anxiety compared to sertraline and citalopram. The latter finding, that fluoxetine was associated with higher odds of anxiety compared to other drugs, is inline with existing literature showing fluoxetine to have a higher likelihood of anxiety as a side effect compared to other SSRIs (Ferguson, 2001). The former finding of paroxetine and fluvoxamine being associated with higher odds of anxiety, however, has not been previously identified (Ferguson, 2001). Additionally, paroxetine and fluoxetine also showed higher odds of searches for depression compared to sertraline and fluoxetine. This finding could represent evidence toward decreased efficacy of paroxetine and fluoxetine in treating depression, though, again, this is not consistent with existing literature demonstrating near equivalent efficacy of SSRIs. Perhaps more likely is confounding by indication, with fluoxetine and sertraline used more commonly to treat depression than fluvoxamine and paroxetine.

Regarding physical symptoms, fluvoxamine, citalopram, and paroxetine were most often associated with higher odds for search queries related to such symptoms. Fluvoxamine, for example, was associated with higher odds of searches for weight gain compared to all other SSRIs. Conversely, escitalopram was associated with lower subsequent searches for weight gain when compared to citalopram, fluoxetine, fluvoxamine, and paroxetine. This finding is generally consistent with escitalopram’s more favorable clinical side effect profile, owing to its more specific molecular binding profile, compared to other antidepressants (Owens et al., 2001). However, while SSRIs as a class are known to cause weight gain (Fava, 2000), significant differences with regard to weight gain between SSRIs are not well established (Blumenthal et al., 2014), with limited support for paroxetine (Fava, 2000; Ferguson, 2001) being worse than the others.

Notably, our results also demonstrated citalopram was associated with drastically higher odds of impotence searches, compared to all other SSRIs, except fluvoxamine. While existing literature shows sexual side effects as common across all SSRIs (Rothschild, 2000), there is some evidence supporting paroxetine, sertraline, and citalopram as worse than the others, with paroxetine commonly used off-label for treating premature ejaculation (Zhang et al., 2019) owing to its delay of orgasm. We additionally found paroxetine was most frequently associated with fatigue compared to other SSRIs. This finding is generally in line with existing literature reporting higher levels of fatigue among those taking paroxetine compared to other SSRIs (Marks et al., 2008). Lastly, sertraline was associated with lower subsequent searches for diarrhea, when compared to citalopram and fluoxetine. This finding is at odds with existing literature, which suggests that sertraline actually has the highest potential for GI side effects when compared to other SSRIs (Wang et al., 2022). Such opposing findings could be due to methodological differences in our data collection compared to more traditional methods for side effect detection; traditional methods often require a patient to not only be consciously aware of a drug effect, but also to correctly attribute that effect to the drug and to report the effect to the proper entity. Unsurprisingly, evidence supports varying side effect prevalence estimates depending on the method used to elicit information from patients (Kelly et al., 2008; Rief et al., 2011).

With regard to topic searches, the SSRIs showed distinct differences in the topics: health, commerce, images, recipes, technology, and celebrities. Specifically, we found that fluvoxamine was associated with higher subsequent searches for health compared to all other SSRIs. One potential explanation is confounding by indication, as fluvoxamine is approved and often used to treat patients with obsessive compulsive disorder (OCD) (Mundo et al., 2001), which is frequently characterized by health-related obsessions. A second striking example was paroxetine’s significant association with later image searches, compared to other SSRIs. While definitive explanations are limited by the type of data available and beyond the scope of the present work, one could hypothesize that image searches may require lower active engagement and appeal as a passive form of entertainment, perhaps related to paroxetine’s greater association with sedation compared to other SSRIs (Supplementary Reference 2).

Results from the current work have important clinical and research implications. Combined with other sources of clinical and pharmaceutical data, the significant associations identified in this study may be used to ground the oftentimes trial and error approach of SSRI prescription in a more empirical foundation. Citalopram, for instance, might be more cautiously prescribed in someone for whom sexual side effects are of particular concern and fluvoxamine might be avoided in someone for whom avoiding weight gain is a priority. While we are not advocating for the results of this study to be used as a singular point of reference for SSRI selection and prescribing, the results can be situated in the growing body of theoretical and empirical guides for clinical decision support (Stahl, 2004; Zeier et al., 2018). As such, the results identified herein could serve to improve personalized prescription.

From a pharmaceutical research standpoint, the present work indicates a specific use case for postmarket drug surveillance as suggested by Yom-Tov and Gabrilovich (2013)–that is, intraclass drug comparison powered by an unprecedented sample size. In exploring differences within the SSRI drug class, our work provides the potential for uncovering previously undetected differences between same-class drugs. The sample size permitted by using search engine “big data” provides adequate power to detect even subtle, but potentially clinically important, differences between same-class drugs, which would be undetected in smaller RCTs. Further, the use of naturalistic data avoids bias potentially present in self-reported adverse event data. We suggest that a similar methodological approach would be particularly useful with the growing number of medication choices available to prescribers for particular indications; we consider the atypical antipsychotics, for instance, as an area where large scale studies powered to detect subtle differences would be particularly impactful due to the severe nature of common side effects.

Despite the strengths of this research, including a large sample size and naturalistic data, it is important to contextualize our findings by highlighting limitations. First, while evidence supports Bing users as representative of internet users, Bing has a minority US market share (Microsoft, 2024). Second, search queries are imperfect proxies for real world behaviors and experiences. Similarly, search data is inherently noisy, and it is seldom possible to discern the reason for a person making a web search (e.g., research for self or for another). Even so, prior work has established a precedent for this proxy (Jacobson et al., 2022; Yom-Tov and Gabrilovich, 2013), with multiple studies strongly linking drug search query data with community prescribing patterns (Simmering et al., 2014; Van Haaren et al., 2023); further the magnitude of our sample aids in mitigating the problem of separating signal from noise. Third, utilizing search data as in the present analysis introduces potential confounding variables, including those related to indication. For example, fatigue could be either a side effect of an SSRI or a symptom of progressively worsening depression (American Psychiatric Association, 2013), which the SSRI was intended to treat. Thus, while our approach is not a replacement for randomized clinical trials or cohort studies, it has the unique potential of comparatively assessing otherwise under- or undetected SSRI effects, guiding future clinical trials. Ongoing work may address these limitations and build on the present work by utilizing search data in the context of clinical trials, which link search data to real world observations, such as severity of depression and SSRI prescription. Jacobson et al. (2022), for instance, partially addressed these challenges by linking search queries to mental health screening surveys.

The results of our work show considerable promise in using search data to distinguish between SSRIs; further, our approach carries the distinct advantages of large samples powered to detect even subtle within-class medication differences, as well as longitudinal time spans capable of detecting temporal medication effects. With a rapidly growing variety of psychiatric pharmaceuticals competing for similar indications, “big search data” affords the possibility of constructing detailed medication profiles, allowing for more evidence based and personalized prescribing decisions.

Supplementary Material

1

Funding

MVH is funded by T32 CA134286 from the NIH. NCJ is funded by R01MH123482 from the NIMH.

Footnotes

Declarations of Interest Statement

NCJ has a received grant from Boehringer-Ingelheim. NCJ has edited a book through Academic Press and receives book royalties, and NCJ also receives speaking fees related to his research. MVH has worked as a paid consultant for Artisight and ChatNexus. EYT was previously employed by Microsoft Research.

Publisher's Disclaimer: This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain.

Bibliography

  1. American Psychiatric Association (Ed.), 2013. Diagnostic and Statistical Manual of Mental Disorders: DSM-5, 5th ed. ed. American Psychiatric Association, Washington, D.C. [Google Scholar]
  2. Blumenthal SR, Castro VM, Clements CC, Rosenfield HR, Murphy SN, Fava M, Weilburg JB, Erb JL, Churchill SE, Kohane IS, Smoller JW, Perlis RH, 2014. An Electronic Health Records Study of Long-Term Weight Gain Following Antidepressant Use. JAMA Psychiatry 71, 889–896. 10.1001/jamapsychiatry.2014.414 [DOI] [PMC free article] [PubMed] [Google Scholar]
  3. Boschloo L, Hieronymus F, Lisinski A, Cuijpers P, Eriksson E, 2023. The complex clinical response to selective serotonin reuptake inhibitors in depression: a network perspective. Transl. Psychiatry 13, 19. 10.1038/s41398-022-02285-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  4. Chokor A, Sarker A, Gonzalez G, 2016. Mining the Web for Pharmacovigilance: the Case Study of Duloxetine and Venlafaxine. 10.48550/arXiv.1610.02567 [DOI]
  5. Coupland C, Dhiman P, Morriss R, Arthur A, Barton G, Hippisley-Cox J, 2011. Antidepressant use and risk of adverse outcomes in older people: population based cohort study. The BMJ 343, d4551. 10.1136/bmj.d4551 [DOI] [PMC free article] [PubMed] [Google Scholar]
  6. Fava M, 2000. Weight gain and antidepressants. J. Clin. Psychiatry 61 Suppl 11, 37–41. [PubMed] [Google Scholar]
  7. Ferguson JM, 2001. SSRI Antidepressant Medications: Adverse Effects and Tolerability. Prim. Care Companion CNS Disord. 3, 23317. 10.4088/PCC.v03n0105 [DOI] [PMC free article] [PubMed] [Google Scholar]
  8. Gahr M, Uzelac Z, Zeiss R, Connemann BJ, Lang D, Schönfeldt-Lecuona C, 2015. Linking Annual Prescription Volume of Antidepressants to Corresponding Web Search Query Data: A Possible Proxy for Medical Prescription Behavior? J. Clin. Psychopharmacol. 35, 681–685. 10.1097/JCP.0000000000000397 [DOI] [PubMed] [Google Scholar]
  9. Garbe E, Suissa S, Douros A, 2019. Pharmacoepidemiology, in: Ahrens W, Pigeot I (Eds.), Handbook of Epidemiology. Springer, New York, NY, pp. 1–55. 10.1007/978-1-4614-6625-3_31-1 [DOI] [Google Scholar]
  10. Gelfand JM, Noe MH, 2021. 7 - Pharmacovigilance: Verifying that Drugs Remain Safe, in: Wolverton SE (Ed.), Comprehensive Dermatologic Drug Therapy (Fourth Edition). Elsevier, pp. 54–61.e2. 10.1016/B978-0-323-61211-1.00007-3 [DOI] [Google Scholar]
  11. Hengartner MP, 2022. Evidence-biased Antidepressant Prescription: Overmedicalisation, Flawed Research, and Conflicts of Interest. Springer International Publishing, Cham. 10.1007/978-3-030-82587-4 [DOI] [Google Scholar]
  12. Jacobson NC, Yom-Tov E, Lekkas D, Heinz M, Liu L, Barr PJ, 2022. Impact of Online Mental Health Screening Tools on Help-Seeking, Care Receipt, and Suicidal Ideation and Suicidal Intent: Evidence from Internet Search Behavior in a Large U.S. Cohort. J. Psychiatr. Res. 145, 276–283. 10.1016/j.jpsychires.2020.11.010 [DOI] [PMC free article] [PubMed] [Google Scholar]
  13. Jakobsen JC, Katakam KK, Schou A, Hellmuth SG, Stallknecht SE, Leth-Møller K, Iversen M, Banke MB, Petersen IJ, Klingenberg SL, Krogh J, Ebert SE, Timm A, Lindschou J, Gluud C, 2017. Selective serotonin reuptake inhibitors versus placebo in patients with major depressive disorder. A systematic review with meta-analysis and Trial Sequential Analysis. BMC Psychiatry 17, 58. 10.1186/s12888-016-1173-2 [DOI] [PMC free article] [PubMed] [Google Scholar]
  14. Jakubovski E, Johnson JA, Nasir M, Müller-Vahl K, Bloch MH, 2019. Systematic review and meta-analysis: Dose-response curve of SSRIs and SNRIs in anxiety disorders. Depress. Anxiety 36, 198–212. 10.1002/da.22854 [DOI] [PubMed] [Google Scholar]
  15. Kelly K, Posternak M, Jonathan EA, 2008. Toward achieving optimal response: understanding and managing antidepressant side effects. Dialogues Clin. Neurosci. 10, 409–418. [DOI] [PMC free article] [PubMed] [Google Scholar]
  16. Kenyon-Mathieu C, Schudy W, 2007. How to rank with few errors, in: Proceedings of the Thirty-Ninth Annual ACM Symposium on Theory of Computing. Presented at the STOC07: Symposium on Theory of Computing, ACM, San Diego California USA, pp. 95–103. 10.1145/1250790.1250806 [DOI] [Google Scholar]
  17. Lewer D, O’Reilly C, Mojtabai R, Evans-Lacko S, 2015. Antidepressant use in 27 European countries: Associations with sociodemographic, cultural and economic factors. Br. J. Psychiatry 207, 221–226. 10.1192/bjp.bp.114.156786 [DOI] [PubMed] [Google Scholar]
  18. Lexapro Package Insert, 2017. [Google Scholar]
  19. Marks DM, Park M-H, Ham B-J, Han C, Patkar AA, Masand PS, Pae C-U, 2008. Paroxetine: safety and tolerability issues. Expert Opin. Drug Saf. 7, 783–794. 10.1517/14740330802423168 [DOI] [PubMed] [Google Scholar]
  20. Microsoft, 2024. Microsoft Search Network Data [WWW Document]. Microsoft Advert. Netw. Search. URL https://about.ads.microsoft.com/en/tools/planning/search-data (accessed 7.28.24). [Google Scholar]
  21. Mundo E, Rouillon F, Figuera ML, Stigler M, 2001. Fluvoxamine in obsessive-compulsive disorder: similar efficacy but superior tolerability in comparison with clomipramine. Hum. Psychopharmacol. Clin. Exp. 16, 461–468. 10.1002/hup.317 [DOI] [PubMed] [Google Scholar]
  22. Murphy SE, Capitão LP, Giles SLC, Cowen PJ, Stringaris A, Harmer CJ, 2021. The knowns and unknowns of SSRI treatment in young people with depression and anxiety: efficacy, predictors, and mechanisms of action. Lancet Psychiatry 8, 824–835. 10.1016/S2215-0366(21)00154-1 [DOI] [PubMed] [Google Scholar]
  23. OECD, 2021. Health at a Glance 2021: OECD Indicators, Health at a Glance. OECD. 10.1787/ae3016b9-en [DOI] [Google Scholar]
  24. Owens MJ, Knight DL, Nemeroff CB, 2001. Second-generation SSRIs: human monoamine transporter binding profile of escitalopram and R-fluoxetine. Biol. Psychiatry 50, 345–350. 10.1016/s0006-3223(01)01145-3 [DOI] [PubMed] [Google Scholar]
  25. Panek E, Nardis Y, Auverset L, 2018. It’s All about Me (or Us): Facebook Post Frequency & Focus as Related to Narcissism. J. Soc. Media Soc. 7, 1–17. [Google Scholar]
  26. Pappa D, Stergioulas LK, 2019. Harnessing social media data for pharmacovigilance: a review of current state of the art, challenges and future directions. Int. J. Data Sci. Anal. 8, 113–135. 10.1007/s41060-019-00175-3 [DOI] [Google Scholar]
  27. Pratt LA, Brody DJ, Gu Q, 2017. Antidepressant Use Among Persons Aged 12 and Over:United States,2011–2014. NCHS Data Brief 1–8. [PubMed] [Google Scholar]
  28. Rief W, Barsky AJ, Glombiewski JA, Nestoriuc Y, Glaesmer H, Braehler E, 2011. Assessing general side effects in clinical trials: reference data from the general population. Pharmacoepidemiol. Drug Saf. 20, 405–415. 10.1002/pds.2067 [DOI] [PubMed] [Google Scholar]
  29. Rosenblum S, Yom-Tov E, 2017. Seeking Web-Based Information About Attention Deficit Hyperactivity Disorder: Where, What, and When. J. Med. Internet Res. 19, e6579. 10.2196/jmir.6579 [DOI] [PMC free article] [PubMed] [Google Scholar]
  30. Rothschild AJ, 2000. Sexual side effects of antidepressants. J. Clin. Psychiatry 61 Suppl 11, 28–36. [PubMed] [Google Scholar]
  31. Saha K, Sugar B, Torous J, Abrahao B, Kıcıman E, De Choudhury M, 2019. A Social Media Study on the Effects of Psychiatric Medication Use. Proc. Int. AAAI Conf. Weblogs Soc. Media Int. AAAI Conf. Weblogs Soc. Media 13, 440–451. [PMC free article] [PubMed] [Google Scholar]
  32. Saha K, Torous J, Kiciman E, Choudhury MD, 2021. Understanding Side Effects of Antidepressants: Large-scale Longitudinal Study on Social Media Data. JMIR Ment. Health 8, e26589. 10.2196/26589 [DOI] [PMC free article] [PubMed] [Google Scholar]
  33. Simmering JE, Polgreen LA, Polgreen PM, 2014. Web search query volume as a measure of pharmaceutical utilization and changes in prescribing patterns. Res. Soc. Adm. Pharm. 10, 896–903. 10.1016/j.sapharm.2014.01.003 [DOI] [PubMed] [Google Scholar]
  34. Song M, Xie Q, 2020. Characterizing the psychiatric drug responses of Reddit users from a socialomics perspective. J. Informetr. 14, 101056. 10.1016/j.joi.2020.101056 [DOI] [Google Scholar]
  35. Stahl SM, 2004. Selectivity of SSRIs: individualising patient care through rational treatment choices. Int. J. Psychiatry Clin. Pract. 8, 3–10. 10.1080/13651500410005487 [DOI] [PubMed] [Google Scholar]
  36. Taliaz D, Spinrad A, Barzilay R, Barnett-Itzhaki Z, Averbuch D, Teltsh O, Schurr R, Darki-Morag S, Lerer B, 2021. Optimizing prediction of response to antidepressant medications using machine learning and integrated genetic, clinical, and demographic data. Transl. Psychiatry 11, 381. 10.1038/s41398-021-01488-3 [DOI] [PMC free article] [PubMed] [Google Scholar]
  37. Van Haaren PCF, Tijdink J, Gerritse FL, 2023. Web Search Query Volume Correlates With Prescription Volumes of Antidepressants and Antipsychotics in the Netherlands and United Kingdom: An Explorative Study. J. Clin. Psychopharmacol. 43, 220. 10.1097/JCP.0000000000001690 [DOI] [PubMed] [Google Scholar]
  38. Wang Z, Li H, Kang Y, Liu Y, Shan L, Wang F, 2022. Risks of Digestive System Side-Effects of Selective Serotonin Reuptake Inhibitors in Patients with Depression: A Network Meta-Analysis. Ther. Clin. Risk Manag. 18, 799–812. 10.2147/TCRM.S363404 [DOI] [PMC free article] [PubMed] [Google Scholar]
  39. White RW, Tatonetti NP, Shah NH, Altman RB, Horvitz E, 2013. Web-scale pharmacovigilance: listening to signals from the crowd. J. Am. Med. Inform. Assoc. 20, 404–408. 10.1136/amiajnl-2012-001482 [DOI] [PMC free article] [PubMed] [Google Scholar]
  40. White RW, Wang S, Pant A, Harpaz R, Shukla P, Sun W, DuMouchel W, Horvitz E, 2016. Early identification of adverse drug reactions from search log data. J. Biomed. Inform. 59, 42–48. 10.1016/j.jbi.2015.11.005 [DOI] [PubMed] [Google Scholar]
  41. Yom-Tov E, Gabrilovich E, 2013. Postmarket Drug Surveillance Without Trial Costs: Discovery of Adverse Drug Reactions Through Large-Scale Analysis of Web Search Queries. J. Med. Internet Res. 15, e124. 10.2196/jmir.2614 [DOI] [PMC free article] [PubMed] [Google Scholar]
  42. Zeier Z, Carpenter LL, Kalin NH, Rodriguez CI, McDonald WM, Widge AS, Nemeroff CB, 2018. Clinical Implementation of Pharmacogenetic Decision Support Tools for Antidepressant Drug Prescribing. Am. J. Psychiatry 175, 873–886. 10.1176/appi.ajp.2018.17111282 [DOI] [PMC free article] [PubMed] [Google Scholar]
  43. Zhang D, Cheng Y, Wu K, Ma Q, Jiang J, Yan Z, 2019. Paroxetine in the treatment of premature ejaculation: a systematic review and meta-analysis. BMC Urol. 19, 2. 10.1186/s12894-018-0431-7 [DOI] [PMC free article] [PubMed] [Google Scholar]

Associated Data

This section collects any data citations, data availability statements, or supplementary materials included in this article.

Supplementary Materials

1
NIHMS2017855-supplement-1.docx (173.8KB, docx)

RESOURCES