Testosterone Science

Research Summaries

Low Testosterone on Social Media: Application of Natural Language Processing to Understand Patients’ Perceptions of Hypogonadism and Its Treatment

Population: Men

Citation

Low Testosterone on Social Media: Application of Natural Language Processing to Understand Patients’ Perceptions of Hypogonadism and Its Treatment

Osadchiy V, Jiang T, Mills JN, Eleswarapu SV.

J Med Internet Res. 2020;22(10):e21383. Published October 7, 2020.

PMID: 33026354 DOI: 10.2196/21383

Study at a Glance

Design
Observational analysis of public posts, using archived Reddit discussions and prospectively collected tweets.
Population
12,665 original Reddit posts, 186,670 replies, and 6,659 tweets. The number of individual contributors, their ages, and their clinical diagnoses were unknown.
Duration
Reddit: December 2015 to May 2019 (42 months). Twitter: June to September 2019 (4 months).
Material analyzed
Public discussions of low testosterone and TRT. The study assigned no treatment.
Outcomes
Recurring discussion topics and the percentage of posts containing selected words.

Question Addressed

What do people ask or say about low testosterone and testosterone replacement therapy (TRT) on Reddit and Twitter? Which of those topics might clinicians raise when a patient comes in for an evaluation?

Plain Language Summary

On Reddit, people asked what their blood test results meant, which specialist to see, and how to take testosterone. About 1 in 4 original posts mentioned a doctor, compared with 1 in 11 replies. On Twitter, people discussed symptoms, changes they attributed to treatment, and possible heart risks. Some also used “low T” as an insult. These findings describe what people posted. The researchers could not verify who had low testosterone or who received treatment [5].

For someone bringing a test result to a visit, a useful starting question is “What have you read about this result that you would like to go over?” A clinician can then explain how symptoms and repeat morning blood tests help establish a diagnosis [1,4,5].

Methods

  • Reddit. The team retrieved original posts and replies from r/Testosterone. Original posts had to contain more than 20 words and replies more than 5 words.
  • Twitter. The team searched for “low testosterone,” “low T,” and “testosterone replacement.” It excluded retweets and tweets containing 5 words or fewer.

The researchers analyzed the 3 datasets separately. After filtering out common words, they used principal component analysis to identify words that tended to occur together, then named each cluster. A separate analysis counted words related to medical visits, cardiovascular risk, and prostate cancer.

Key Findings

  • Reddit discussions. Original posts and replies shared 4 themes: seeing a doctor, interpreting test results, administering TRT, and lifestyle changes. Symptoms formed a 5th theme in original posts. Medications, including clomiphene and human chorionic gonadotropin, formed the 5th theme in replies.
  • Finding a clinician. “Doctor” appeared in 3,235 of 12,665 original Reddit posts (25.54%), 9.00% of replies, and 1.77% of tweets. “Urologist” appeared in 5.78%, 1.36%, and 0.20%, respectively.
  • Twitter discussions. The 4 themes were symptoms, cardiovascular risk, perceived symptom improvement, and insults. The word “heart” occurred in 1.14% of tweets. Prostate cancer did not emerge as a distinct theme on either platform, although “prostate” and “cancer” each appeared in 1.13% of original Reddit posts.
  • Statistical results. The clusters explained 15.45% of variation in word use in original posts, 13.84% in replies, and 9.01% in tweets. Bartlett’s test supported factor analysis in all 3 datasets (P<0.01). The paper reported descriptive frequencies without 95% confidence intervals or adjusted clinical effect estimates.

Limitations and Strengths Affecting Interpretation

Strengths

  • Public posts captured questions people raised in their own words.
  • Analyzing opening posts and replies separately showed which topics arose when people asked for help and which arose in responses.

Limitations

  • Contributors chose to post and could post repeatedly. Their ages, diagnoses, and treatment histories were unknown.
  • Word clusters could identify discussion of heart risk but could not show whether a writer endorsed or disputed a claim.

Use these findings to identify questions patients may bring to a visit. Estimating how common those concerns are requires data from a representative sample of patients.

Authors’ Conclusions

The authors suggested asking patients what they have encountered online and inviting them to discuss questions about symptoms, test results, and treatment during the consultation (Osadchiy 2020).

Commentary

An earlier clinic survey of 382 men found that cost influenced the choice of injections and convenience influenced the choice of pellets. The Reddit analysis extends that observation by showing what people asked peers about giving injections, choosing a schedule, and interpreting laboratory results between visits (Kovac 2014, Osadchiy 2020).

A later UK survey found that 81% of 905 men receiving treatment relied on online sources for information. At consultations, 77% wanted to discuss diagnosis and 51% wanted to learn about side effects. A clinician might therefore ask which result or treatment claim prompted the visit and address that concern first (El-Osta 2025, Osadchiy 2020).

For a patient who brings in a single testosterone result, the Endocrine Society recommends reviewing compatible symptoms and confirming the result with a repeat fasting morning measurement. The American Urological Association also calls for 2 early-morning measurements on separate occasions. Explaining the purpose of the second sample directly addresses a recurring question in the Reddit discussions (Bhasin 2018, Mulhall 2018, Osadchiy 2020).

Several features limit how clinicians can interpret the findings (Osadchiy 2020).

  • Selection. A dedicated forum and a search limited to 3 Twitter phrases captured different audiences over different periods. The results cannot establish which concerns are most common in a clinic population (Osadchiy 2020).
  • Meaning. Word clusters can identify a discussion about heart risk but cannot establish whether the writer feared a risk, disputed it, or was repeating a headline (Osadchiy 2020).
  • Clinical outcomes. Treatment histories and changes in symptoms could not be independently verified (Osadchiy 2020).

Interpret this study as an account of the questions and concerns expressed in the sampled online discussions. Clinically, the findings support asking patients what they have read online, reviewing their understanding of laboratory results, and discussing treatment expectations and safety concerns during the consultation. Applying these findings to communication should accompany diagnostic assessment based on symptoms and repeat morning testosterone measurements. The frequency of a word in a post cannot establish how common a concern is among patients or estimate treatment benefit or harm (Bhasin 2018, Mulhall 2018, Osadchiy 2020).

References

1. Bhasin S, Brito JP, Cunningham GR, Hayes FJ, Hodis HN, Matsumoto AM, Snyder PJ, Swerdloff RS, Wu FC, Yialamas MA. Testosterone therapy in men with hypogonadism: an Endocrine Society clinical practice guideline. J Clin Endocrinol Metab. 2018;103(5):1715-1744.

| PMID

Supporting quote: “We recommend confirming the diagnosis by repeating the measurement of morning fasting total T concentrations.”

2. El-Osta A, Liu VN, Huang D, Foster J, Johnson H, Alaa A, Hayhoe B. A cross-sectional survey of experiences and outcomes of using testosterone replacement therapy in UK men. Transl Androl Urol. 2025;14(5):1295-1307.

| PMID

Supporting quote: “Most (81%) relied on online sources for information”

3. Kovac JR, Rajanahally S, Smith RP, Coward RM, Lamb DJ, Lipshultz LI. Patient satisfaction with testosterone replacement therapies: the reasons behind the choices. J Sex Med. 2014;11(2):553-562.

| PMID

Supporting quote: “Convenience and ease of use are central in choosing pellet therapy.”

4. Mulhall JP, Trost LW, Brannigan RE, Kurtz EG, Redmon JB, Chiles KA, Lightner DJ, Miner MM, Murad MH, Nelson CJ, Platz EA, Ramanathan LV, Lewis RW. Evaluation and management of testosterone deficiency: AUA guideline. J Urol. 2018;200(2):423-432.

| PMID

Supporting quote: “accurate assessment of total testosterone levels, symptoms, and signs”

5. Osadchiy V, Jiang T, Mills JN, Eleswarapu SV. Low Testosterone on Social Media: Application of Natural Language Processing to Understand Patients’ Perceptions of Hypogonadism and Its Treatment. J Med Internet Res. 2020;22(10):e21383.

| PMID

Supporting quote: “Over a quarter of Reddit parent posts contain the term doctor”

Original article © 2020 the authors. Published by JMIR under.

All Research Summaries

Post a Comment or Question

Shared Decision Making

Bring the Evidence to Your Next Consultation

Use our patient resources and questionnaires to prepare questions about symptoms, testing and treatment options for your doctor.