The placebo problem in pain: does baseline pain variability predict placebo response?

May 11, 2022

Recent reports have demonstrated a positive correlation between baseline pain variability and improvement in clinical trial patients receiving placebo, suggesting that patients with more highly variable reporting of clinical pain can be expected to have higher placebo response. For example, a 2019 paper by Triester, et al. examined this relationship in a clinical trial evaluating the efficacy of 2 week treatment with naproxen for osteoarthritis pain in a crossover design1. This study found a statistically significant correlation between variability in daily pain reporting for 1 week at baseline (i.e. before randomization) and clinical improvement in placebo-treated patients (Pearson’s correlation = 0.393; p=0.04)1. These results were consistent with those reported by others using a logistic regression model to identify responders (defined as patients experiencing a 30% reduction in pain after placebo treatment). Bonin Pinto, et al. found that groups displaying higher variability in pain reporting also presented larger improvements in pain in knee osteoarthritis (Odds ratio 1.6-2.3)2; while Ballou, et al. found that baseline pain variability was associated with placebo response in patients with irritable bowel syndrome (IBS) with constipation (odd ratio = 1.71)3.

Despite these compelling reports, this topic remains controversial in the literature. First, other published studies have also failed to find a relationship between baseline pain variability and placebo response. For example, a 2022 publication by Gillving, et al. reported no relationship between in a meta-analysis of 3 clinical trials in peripheral neuropathic pain using crossover designs4. It must also be recognized that the statistical phenomenon of regression to the mean can also be interpreted as placebo response5, and also that greater symptomatic variability increases the likelihood of regression to the mean. Indeed, in a letter to the editor, Dr. Robert Palmer suggested that regression to the mean – and not placebo response per se – may be partly or largely responsible for the associations published by Ballou, et al. in IBS with constipation6. Furthermore, Tiwari, et al. recently compared placebo treatment with a no-treatment group to control for regression to the mean and the natural disease history in two independent clinical trial in chronic low back pain7. The results from this analysis suggested that baseline pain variability accounted for <1% of the variance in post-intervention pain across both studies and concluded that more work needed to be done before pain variability could considered to be consistently and reliably prognostic for placebo response. 

Further complicating this issue, pain reporting variability is not necessarily consistent over time. In “Can Daily Self-Assessment Induce A Learning Effect Mitigating Pain Evaluation Error In Randomized Clinical Trials?” by Ooghe, et al., pain reporting variability, measured as the Pearson’s auto-correlation of daily pain measurements between two successive days, was evaluated for clinical trial patients with hip and/or knee OA for 12 weeks8. The variability in reporting of all daily pain measurements (average pain, worse pain, and lowest pain) decreased throughout the trial, as the correlation between measurements was significantly higher after 12 weeks than at the beginning of the study8. This suggests that simply reporting pain daily is itself a viable method to reduce pain reporting variability over time, and that baseline pain variability is not consistent throughout the length of a typical clinical trial.

At Cognivia, our interest in better understanding the placebo response in pain has motivated us to also investigate the importance of baseline pain reporting variability. In “Impact of excluding highly variable pain subjects on the treatment estimation”, data from a clinical trial in patients with moderate to severe knee osteoarthritis were analyzed to understand the relationship between pain reporting variability, pain improvement and assay sensitivity. These data demonstrated that pain reporting variability is weakly related to pain improvement in a clinical trial, with R2 of 4.4%. As such, in this study, pain reporting variability can explain less than 5% of data variability related to the placebo response.  

While baseline pain variability may weakly contribute to individual patient placebo responsiveness, the data generated with the Placebell©™ method suggest that other factors contribute far more significantly. In the development of Placebell©™, machine learning-based predictive models have been built and validated in chronic pain that have identified factors that robustly and reproducibly contribute to the prediction of placebo response and the distinction between placebo responders and non-responders. These factors include patient psychological traits, expectation, baseline disease intensity and demographics, among others.  In contrast to the ~4% of data variability that is related to pain reporting, Placebell©™ has reproducibly explain 30% or more variability in placebo responsiveness on a per patient basis in chronic pain using a pre-specified algorithm9. Furthermore, Placebell©™ has been demonstrated to improve assay sensitivity – or the ability of distinguish placebo from drug treatment – by nearly 40%9. On the other hand, reducing baseline pain variability in the clinical trial patient population by excluding patients with the greatest variability has no net effect on assay sensitivity10

In summary, while individual patient pain reporting variability may have limited prognostic value in understanding placebo responsiveness, the Placebell©™ approach – which considers multiple patient factors in a specific combination as defined by machine learning-based modeling – has been proven to reduce data variability related to the placebo response and improve assay sensitivity.  For more information about how Placebell©™ can applied in clinical trials in pain or other indications, please contact us.


1.        Treister R, Honigman L, Lawal OD, Lanier RK, Katz NP. A deeper look at pain variability and its relationship with the placebo response: results from a randomized, double-blind, placebo-controlled clinical trial of naproxen in osteoarthritis of the knee. Pain. 2019;160(7):1522-1528. doi:10.1097/J.PAIN.0000000000001538

2.        Bonin Pinto C, Barroso J, Schinitzer TJ. Characterization and implications of daily pain variability and response to treatment in osteoarthritis clinical trials. Osteoarthritis and Cartilage. 2021;29:S277. doi:10.1016/J.JOCA.2021.02.364

3.        Ballou S, Beath A, Kaptchuk TJ, et al. Factors Associated With Response to Placebo in Patients With Irritable Bowel Syndrome and Constipation. Clinical Gastroenterology and Hepatology. 2018;16(11):1738-1744.e1. doi:10.1016/J.CGH.2018.04.009

4.        Gillving M, Demant D, Holbech J v., et al. Impact of variability in baseline pain on the placebo response in randomized, placebo-controlled, crossover trials in peripheral neuropathic pain. Pain. 2022;163(3):483-488. doi:10.1097/J.PAIN.0000000000002374

5.        McDonald CJ, Mazzuca SA, McCabe GP. How much of the placebo “effect” is really statistical regression? Stat Med. 1983;2(4):417-427. doi:10.1002/SIM.4780020401

6.        Palmer RH. Predictors of the Placebo Response in Irritable Bowel Syndrome With Constipation. Clinical Gastroenterology and Hepatology. 2019;17(6):1215-1216. doi:10.1016/J.CGH.2018.12.020

7.        Tiwari SR, Vigotsky AD, Apkarian AV. On the Relationship Between Pain Variability and Relief in Randomized Clinical Trials. Frontiers in Pain Research. 2022;3:844309. doi:10.3389/FPAIN.2022.844309

8.        Ooghe A, Branders S, Pereira A. Can daily self-assessment induce a learning effect mitigating pain evaluation error in clinical trials? Osteoarthritis and Cartilage. 2021;29:S262-S263. doi:10.1016/j.joca.2021.02.346

9.        Branders S, Dananberg J, Clermont F, et al. Predicting the placebo response in oa to improve the precision of the treatment effect estimation. Osteoarthritis and Cartilage. 2021;29:S18-S19. doi:10.1016/j.joca.2021.05.032

10.      Ooghe A, Branders S, Demolle D, Pereira A. Optimizing assay sensitivity by combining highly variable pain subjects’ exclusion with adjusted analysis. Accessed May 9, 2022.


VP, Business Development

Related content


Leveraging Historical Data For High-dimensional Regression Adjustment, A Machine Learning Approach

Samuel Branders, Ph.D., Data Mining and Statistical Research Scientist of Tools4Patient (T4P) recently presented cutting-edge research at the...

Read More

Compilation Of Frequently Asked Questions About Placebell©™

In May 2018, Tools4Patient (T4P) presented a webinar entitled “Characterization of Individual Patient Placebo Response: Impact on the...

Read More

Predicting The Placebo Response In Chronic Pain Patients

There is substantial scientific literature describing the placebo response in pain, including characterizing the magnitude and duration of...

Read More

Let’s help you with your next trial.

This field is for validation purposes and should be left unchanged.