Abstract

Online reviews offer a wealth of information regarding drug efficacy. They reveal the side effects of treatments, as well as the joy and frustration experienced by real people. In this blogpost, we used natural language processing (NLP) to understand what patients are saying regarding their treatments, using birth control methods as a specific example.

Introduction

In the United States, 65% of women of childbearing potential use at least one method of contraception. With more than a dozen options in the marketplace, which birth control works best? In this blogpost, we will try to answer this question with an innovative approach, drawing insights from thousands of women who have shared their experiences online. Textual information from reviews was distilled via natural language processing (NLP), a branch of machine learning that seeks to understand written and spoken language. The applications of NLP include conversational Artificial Intelligence such as Siri, but here we will use it to understand what patients are saying online.

Data Sources

Data from this blogpost came from Drugs.com. Drugs.com is a popular, comprehensive website that provides health information. It’s like Wikipedia for drugs! A feature of this website is that anyone can share their experiences by posting a review. Here is an example of a review for Mirena (a brand-name contraceptive):

These online reviews offer a rich amount of information. They reveal the efficacy, side effects, and overall user experience of drugs from the point of view of real patients. However, reviews can be difficult to interpret due to the amount of conflicting information. For example, here are two real, opposing reviews about the same treatment:

“I have used Mirena for over a few years, and I love it”

“Do not get the Mirena! It will ruin your life!”

Whose advice should we follow? To make sense of these mixed reviews, we must move past anecdotes and summarize the overall trend.

What are users talking about?

For popular contraceptives such as Mirena (an IUD), there are hundreds of reviews on Drugs.com. It is labor-intensive to read all the reviews and quite impossible to summarize this information using our memory alone. Instead, we will use natural language processing to distill the information.

Here, we used a technique called topic modeling to identify the common themes that emerged from these reviews. Here are the top four popular topics identified from online reviews for Mirena:

Note: bubble size corresponds to popularity of the topic.

Comparing popular contraceptives

From the section above, we saw that many birth control users talked about the side effects they experienced. To determine which birth control works best, we can potentially examine which birth control has the lowest occurrence of side effects. Of course, there are other factors one must consider when choosing a birth control, such as the ease of use or longevity. But for the purpose of this blog post, let’s focus on comparing the side effects profile of different birth controls.

Specifically, we examined three blockbuster birth controls: Mirena, Lo, and Nuvaring. Results are shown in the table below. For example, 60.6% of reviewers for Mirena mentioned they experienced pain and cramping.

Note that sometimes people negate a side effect. For example, someone may say, “Mirena is great! I had no headache, no bleeding.” It’s important that we avoid counting these mentions as reported side effects. To this end, the percentages reported in the table above have been adjusted for negations.

Validity of this approach

Now that we have the numbers, can we trust them? Are online reviews a valid way to understand side effects? It is possible that patients who are more upset are more motivated to post a review, therefore painting a more negative picture than the reality. One way to evaluate the validity of our approach is to compare our results to randomized, controlled clinical trials, which are often considered as the gold standard for conducting medical research. Let’s compare the side effects reported through these two different approaches:

The frequencies of side effects found online are fairly consistent with the frequencies found in clinical trials, with the exception of weight, mood, and libido changes. For these three side effects, the occurrences reported online are much higher.

What is the reason for such a discrepancy? The frequency of side effects reported online may differ from the ones reported in clinical trials for several reasons:

Patients who experienced adverse events may be more motivated to share their experience online, inflating the number of side effects online
Patients who experienced adverse events may drop out early from clinical trials, deflating the number of side effects in clinical trials
Patients recruited in clinical trials may be healthier than the general public, if the clinical trials had excluded patients with prior conditions and other complications.

Overall, we need to be careful with our interpretation of online reviews, since there may be a negativity bias online.

Conclusion

In this blog post, we used natural language processing to understand people’s experiences with different types of birth controls, synthesizing insights from thousands of online reviews. We saw that the frequency of side effects can be different depending on the birth control. Among Mirena users, pain was reported considerably more than among Lo and Nuvaring users, potentially because the IUD insertion itself can be painful. We also observed that mood changes are more common among Lo and Nuvaring users compared to Mirena users.

Without having to scroll through hundreds of reviews, our analysis uncovered personal insights from the online community regarding different types of treatments. By drawing wisdom from the community, this kind of analysis can help people make more informed decisions regarding medical treatments, and we will be expanding these analyses to other therapeutic areas.

References

Amanda López-Picado, Oihane Lapuente & Iñaki Lete. Efficacy and side-effects profile of the ethinylestradiol and etonogestrel contraceptive vaginal ring: a systematic review and meta analysis. Eur J Contracept Reprod Health Care. 2017;22(2):131-146.
Hans-Joachim Ahrendta, Israel Nisandb, Carlo Bastianelli, et al. Efficacy, acceptability and tolerability of the combined contraceptive ring, NuvaRing, compared with an oral contraceptive containing 30 lg of ethinyl estradiol and 3 mg of drospirenone. Contraception. 2006;74(6):451-7.
N. N. Sarkar. The combined contraceptive vaginal device (NuvaRing1): A comprehensive review. The European Journal of Contraception and Reproductive Health Care June 2005;10(2):73–78
Dieben TOM, Roumen FJME, Apter D. Efficacy, cycle control, and user acceptability of a novel combined contraceptive vaginal ring. Obstet Gynecol 2002; 100: 585-93.
Mitchell D Creinin, Leslie A Meyn, Lynn Borgatta, et al. Multicenter comparison of the contraceptive ring and patch: a randomized controlled trial. Obstet Gynecol. 2008;111(2 Pt 1):267-77.
Rosa Sabatinia, Raffaele Cagiano. Comparison profiles of cycle control, side effects and sexual satisfaction of three hormonal contraceptives. Contraception. 2006;74(3):220-3.
Carolina Sales Vieira, Ian S. Fraser, Marlena G Plagianos, et al. Bleeding profile associated with 1-year use of the segesterone acetate/ethinyl estradiol contraceptive vaginal system: pooled analysis from phase 3 trials.Contraception. 2019; 100(6): 438–444.
Kristina Gemzell-Danielsson, Regine Sitruk-Ware, Mitchell D Creinin, et al. Segesterone acetate/ethinyl estradiol 12-month contraceptive vaginal system safety evaluation. Contraception. 201;99(6):323-328.

Which Birth Control is Best?