Placebo Effect Sizes in Clinical Trials of Knee Osteoarthritis Using Intra‐Articular Injections of Biologic Agents

Abstract

Objective

Patients with knee osteoarthritis rely on symptomatic treatments, in which up to 75% of the pain reduction can be attributed to the placebo effect. This effect may vary based on treatment type (eg, biologics vs nonbiologic injection) and route of administration (eg, intra‐articular vs topical vs oral). The placebo effect is an integral part of treatment effect size calculation; thus, network analyses comparing efficacies of different treatments may be inaccurate. The objective of this study was to test the hypothesis that placebo effects differ between treatment types and route of delivery.

Methods

A systematic literature search was conducted in August 2019. Randomized trials comparing pain outcomes of oral, topical, or intra‐articular placebo interventions to active treatments were included. The outcome measure of interest was change in pain scores from baseline. Data were stratified by length of follow‐up and treatment subcategory.

Results

A total of 129 articles were included with 9,218 patients receiving placebo treatments. Reduction in pain from baseline occurred in 93% of the subcategory data points. Biologic intra‐articular placebo injections had the greatest pain reduction at one month (mean ± SD visual analog scale −32.2 ± 24.6; mean ± SD Western Ontario and McMaster Universities Arthritis Index −16.3 ± 3.81). At one month and two months, placebo intra‐articular injections had a greater pain reduction than oral placeboes (P ≤ 0.01).

Conclusion

The robust placebo effect is influenced by the active treatment category and changes over time. The variation in placebo response despite analogous placebo methodologies implies using network meta‐analyses to compare treatments from different active treatment categories by evaluating the change from placebo is inaccurate.

INTRODUCTION

Knee osteoarthritis (KOA) affects approximately 14 million people in the United States alone, ¹ but there are no proven effective disease‐modifying treatments available for this painful condition. Outside of surgical options, patients must rely on weight management, physical therapy, and analgesic treatments, in which up to 75% of the pain reduction can be attributed to the placebo response. ² , ³ This reduction in pain due to the placebo effect is not trivial, with recent studies demonstrating that placebo treatments achieve a clinically meaningful response. ⁴ Although the placebo effect can be significant, it is also highly variable; the degree of placebo effect varies based on many factors including the patient, delivery of care, modality of administration, and cost of the treatment. ⁵ , ⁶ , ⁷

SIGNIFICANCE & INNOVATIONS.

• Our article demonstrates the clinical importance of the placebo effect with the finding that more than 90% of subcategory data points had a significant reduction in pain from baseline and almost half had a pain reduction greater than minimally clinically important difference.

• Intra‐articular saline injections associated with biologics have a significantly greater reduction in pain than intra‐articular saline injections associated with nonbiologics at one month, suggesting the subcategory of active treatment may contribute to the power of the placebo and should be considered when powering a clinical trial.

• The placebo effect changes over time with significant differences being seen in almost all subcategories of active treatments. This calls into question the practice of combining results spanning broad time intervals.

• The variation in placebo response despite analogous placebo methodologies implies meta‐analyses comparing treatments from different active treatment categories by evaluating the change from placebo are inaccurate.

The variation in the therapeutic effect of placebo treatments has become a topic of increasing discussion with the advent of novel biologic treatments for KOA. Questions have been raised if the unconvincing results of intra‐articular biologic injections in randomize controlled trials (RCTs) are due to a more powerful placebo effect instead of a true lack of efficacy in these promising treatments. ⁸ Unlike other therapies, which must go through rigorous testing before being advertised to the public, many biologic treatments for KOA have been exempt from premarket review, falling under Public Health Section Act Section 361 Pathway, a pathway created to expediate the development‐to‐market. ⁹ With this exemption and a marked acceleration in capital investments, there has been rapid growth in demand of these novel intra‐articular injections fueled by advertising and potential financial incentives for practitioners. ⁹ Before any clear evidence to justify their use, hundreds of clinics across the United States began marketing stem cell treatment for orthopedic conditions. ¹⁰ The excitement is not completely unfounded. The proof‐of‐concept evidence from preclinical studies was promising, with animal models demonstrating alterations in cellular signaling pathways leading to significant regeneration potential. ¹¹ However, public enthusiasm can lead to increased response expectancy, one of the known factors to affect the placebo effect. ¹²

It has been argued that the optimistic public expectations of these new treatments could account for the large placebo effect noted in the initial RCTs, reducing the apparent effect size of the active treatment. ⁸ The impact of the variation in placebo response does not end with its impact on between‐group comparisons in individual trials. It is especially relevant for network meta‐analyses, a crucial part of creating guidelines that best serve patients, in which effect sizes are directly compared based on the assumption of transitivity. There has been a surge of network meta‐analyses for KOA treatments published over the last five years, with a massive increase in those comparing biologics to nonbiologic intra‐articular injections in the last year. ¹³ , ¹⁴ , ¹⁵ , ¹⁶ , ¹⁷ No previous study has assessed if the pain reduction due to the placebo response is equivalent between the subcategory of active treatment (ie, biologics vs nonbiologic), which is vital information for these studies to be valid. Previous studies have found varying reduction in pain based on the category of placebo (ie, oral vs topical vs intra‐articular), ⁶ and we hypothesized that the subcategory of active treatment may also influence the effectiveness of the intra‐articular placebo, as well as the other main categories of KOA treatment. These differing placebo responses need to be investigated to fairly assess new treatments and to accurately compare treatment options from different subcategories in network meta‐analyses. In summary, the aim of this study was to investigate whether placebo effects on pain reduction differ in KOA treatments depending on treatment types and route of administration.

MATERIALS AND METHODS

Retrieval of published studies

A systematic literature search was conducted in July 2019 using PubMed, Embase, and Google Scholar from the inception date of each database. Search criteria is detailed in Supplementary Table 1, and the search criteria per subcategory (ie, “KOA,” “outcome,” “treatment,” and “study”) were linked using the AND function. There were no language restrictions for this search. We manually searched reference lists for all Cochrane review articles to identify any additional relevant studies for inclusion. Citations and abstracts retrieved from this search were downloaded to EndNote and duplications were removed ¹⁸ (Figure 1).

Figure 1.

Flow diagram for study selection. VAS, visual analog scale; WOMAC, Western Ontario and McMaster Universities Arthritis Index.

Inclusion and exclusion criteria

We included randomized, placebo‐controlled trials including crossover trials of adults diagnosed with KOA. Included trials are listed and cited under Supplementary Table 2. Records were included if they compared the pain outcomes in either Western Ontario and McMaster Universities Arthritis Index (WOMAC) and visual analog scale (VAS) of a placebo intervention to at least one active comparer (Figure 1). Full‐text articles were included if placebo interventions were oral placebo, topical placebo, and intra‐articular placebo. Studies investigating other interventions, such as orthoses, acupuncture, exercise, surgery, and laser therapies were excluded (Figure 1). The categories of included nonplacebo active treatments were oral medications, oral supplements, biologic intra‐articular injections, nonbiologic intra‐articular injections, synthetic topical creams, and naturopathic topical creams. Trials including other joints were excluded. Articles that did not report outcome data for any time points between four weeks and one year from baseline were excluded along with measurements that were taken after crossover. Two reviewers (JMB and TS) independently screened titles and abstracts identified by the searches for eligibility using Rayyan Software. ¹⁹ Subsequently, two reviewers (IP and JMB) independently screened full articles. Any disagreements were resolved by consensus.

Figure 2.

Placebo effect for WOMAC pain categorized by subcategory of active treatment. Data are mean and 95% confidence intervals from individual meta‐analyses per time point (Supplementary Figure 1). MCID, minimally clinically important difference; WOMAC, Western Ontario and McMaster Universities Arthritis Index.

Data extractions

A data extraction form was created, tested on 12 randomly selected studies, and refined as necessary. Two reviewers (TS and IP) independently extracted data including last name of the first author, year of publication, location, study inclusion criteria, population characteristics, treatments, outcome measures, length of follow‐up, risk of bias, and the results of each study. We referred to Cochrane's handbook for extracting study results and result presentation (Chapter 7.7). ²⁰ Discrepancies were discussed and ratified by a third assessor (JMB). The outcome measure of interest was change in pain scores from baseline. Given the scarcity of VAS and WOMAC follow‐up data at over one year, this study included data points between four weeks and one year. Intention‐to‐treat data were used when available.

Quality assessment

Trial quality was assessed using a modified version of the Cochrane Risk Bias tool. ²⁰ Bias from random sequence generation, allocation concealment, baseline values, participant masking, assessor masking, outcome data completeness, and selective reporting were all graded individually on a low, medium, or high scale. Use of intention‐to‐treat analysis and industry funding were also recorded.

Data synthesis

If WOMAC and VAS data were reported on a scale other than 1 to 100 points, then the data were normalized to a 1‐ to 100‐point scale and grouped by the length of follow‐up and active treatment category (intra‐articular injection, oral, and topical) and subcategory. Subcategories included nonbiologic intra‐articular injections (eg, hyaluronic acid), biologics intra‐articular injections (eg, platelet‐rich plasma), oral supplements (eg, garlic tablets) and oral medication (eg, acetaminophen), topical naturopathic cream (eg, mud gel), and topical synthetic cream (eg, nonsteroidal anti‐inflammatory drug [NSAID] gels). Categorizations were made by JMB and ratified by KK. If a study reported both VAS and WOMAC scores, then both scores were extracted and analyzed separately. WOMAC results were plotted against a minimally clinically important difference (MCID) of −7.5, representing a 1.5 of 20 pain score decrease, and VAS results against an MCID ⁷ of 13.7.

Statistical analysis

Classical meta‐analyses were performed per treatment and time point using a DerSimonian‐Laird random effects model in JASP (version 0.14.1; JASP team 2020). ²¹ Random effects and 95% confidence intervals (CIs) were compared using an ordinary one‐way analysis of variance (ANOVA) with Tukey's post hoc test per time point. If there were only two groups per time point, an unpaired t‐test with Welch's correction was performed in Prism (version 9.0.1; GraphPad Software LLC.). The nature of the currently available studies investigating KOA did not allow a two‐way ANOVA (group × time) because of too many missing time points and different number of studies available per meta‐analysis. Data were extracted from previously published studies. Therefore, local approval by the institutional review board was not required.

RESULTS

Characteristics of trials

After duplicates were removed from the initial 2,019 records initially identified, 1,314 records were initially screened for eligibility. A total of 129 trials were included in the meta‐analysis (Figure 1). Two‐thirds of the 32,391 total patients in these studies were female. The mean ± SD age was 61.6 ± 10.3 years, and the mean ± SD body mass index was 30.3 ± 6.5. 9218 (28.5% of total patients in the studies received placebo treatments) (Table 1, Supplementary Table 2). There were 27 nonbiologic intra‐articular injection articles with a total of 1,741 participants who received a placebo treatment, 10 biologic intra‐articular injection articles with a total of 471 participants who received a placebo treatment, 26 oral supplement articles with a total of 989 participants who received a placebo treatment, 52 oral medication articles with a total of 4,751 participants who received a placebo treatment, 3 naturopathic topical cream articles with a total of 73 participants who received a placebo treatment, and 11 synthetic topical cream articles with a total of 1,159 participants who received a placebo treatment. All subcategories included more women than men (Table 2, Supplementary Table 2). Only one study did not allow for rescue medication. The only subcategory in which the majority of studies did not have a specified limit for rescue medication use was naturopathic topical creams, in which two of the three studies did not report on rescue medication. The risk of bias in the treatment group was considered relatively high as 69.8% of studies reported industry funding (Supplementary Table 3). Furthermore, in 19.4% to 27.1% of studies, there were some concerns related to assessor masking, participant masking, and/or allocation concealment (Supplementary Table 3).

Table 1.

Demographic data for included studies^*

Characteristic	Data
Total number of studies, N	129
Total patients, N	32,391
Female, n (%) a	21,582 (66.6)
Age, mean (SD) a	61.6 (10.3)
BMI, mean (SD) a	30.3 (6.5)
Rescue medications, n (%)
Not reported	31 (24.0)
Specified limit	71 (55.0)
Unspecified amount	26 (20.2)
Not allowed	1 (0.8)
Placebo patients, n (%)	9,218 (28.5)
Placebo baseline, mean (SD)
VAS	61.3 (15.5)
WOMAC	52.5 (15.4)

^*BMI, body mass index; VAS, visual analog scale; WOMAC, Western Ontario and McMaster Universities Arthritis Index.

^a Sex: n = 1 (0.7%) missing data; age: n = 11 (8.5%) missing data; BMI: n = 54 (41.8%) missing data; baseline n = 11 (8.5%) missing data.

Table 2.

Demographics by subcategory^*

Characteristic	Intra‐articular injection		Oral pill		Topical cream
	Nonbiologic	Biologic	Supplement	Medication	Naturopathic	Synthetic
Total number of studies, N	27	10	26	52	3	11
Total patients, N	4,440	1,411	2,614	20,708	146	2,973
Female, n (%) a	2,837 (63.9)	803 (56.9)	1,763 (67.4)	14,030 (67.8)	131 (89.7)	1,919 (64.5)
Age, mean (SD) a	63.0 (5.4)	58.9 (10.1)	56.2 (9.1)	61.8 (10.8)	55.8 (9.2)	64.5 (9.5)
BMI, mean (SD) a	30.1 (5.7)	29.9 (6.3)	26.9 (4.0)	30.8 (7.1)	28.0 (5.0)	30.6 (5.9)
Rescue medications, n (%)
Not reported	6 (22.2)	3 (30.0)	10 (38.5)	9 (17.3)	2 (66.7)	1 (9.1)
Specified limit	13 (48.1)	4 (40.0)	13 (50.0)	31 (59.6)	1 (33.3)	8 (72.7)
Unspecified amount	8 (29.6)	2 (20.0)	2 (7.7)	12 (23.1)	0 (0.0)	2 (18.2)
Not allowed	0 (0.0)	0 (0.0)	1 (3.8)	0 (0.0)	0 (0.0)	0 (0.0)
Placebo patients, n (%)	1,741 (39.2)	471 (33.4)	989 (37.8)	4,751 (22.9)	73 (50.0)	1,159 (39.0)
Placebo baseline, mean (SD)
VAS	57.5 (18.6)	59.6 (15.1)	51.4 (17.2)	62.1 (14.2)	73.0 (7.8)	69.8 (15.0)
WOMAC	53.4 (14.2)	42.2 (13.5)	53.8 (15.9)	52.2 (16.5)	57.4 (20.5)	54.7 (13.7)

^*BMI, body mass index; VAS, visual analog scale; WOMAC, Western Ontario and McMaster Universities Arthritis Index.

^a Sex: among medication articles, n = 1 (4%) missing data; age: among nonbiologic articles, n = 3 (11%) missing data, among supplement articles, n = 2 (8%) missing data, among medication, n = 4 (8%) missing data, among synthetic articles, n = 1 (9%) missing data, among naturopathic articles, n = 1 (33%) missing data; BMI: among nonbiologic articles n = 13 (48.1%) missing data, among biologic n = 1 (10%) missing data, among supplement articles, n = 8 (31%) missing data, among medication articles, n = 22 (42%) missing data, among synthetic articles, n = 6 (55%) missing data, among naturopathic articles, n = 2 (67%) missing data; baseline: among nonbiologic articles n = 2 (7.4%) missing data, among biologic articles n = 2 (20%) missing data, among medication articles n = 7 (13.5%) missing data.

Pain reduction

Sixty‐eight (93%) of the 78 subcategory data points had a reduction in pain from baseline. The exceptions were naturopathic placebo WOMAC scores at 1 and 1.5 months, biologic VAS score at 1.5 months, medication VAS score at 5 months, and nonbiologic VAS score at 12 months. Thirty‐six data points (49.3%) had a pain reduction greater than MCID. At one month, the biologic intra‐articular placebo injection had the greatest mean pain reduction (mean ± SD VAS −32.2 ± 24.6; mean ± SD WOMAC −16.3 ± 3.81). The WOMAC pain reduction associated with biologic intra‐articular placebo injection was significantly greater than nonbiologics intra‐articular injections (P < 0.0001), oral medication (P < 0.0001), oral supplements (P < 0.0003), and naturopathic creams (P < 0.0154). There was no significant change from one‐month mean pain reduction for biologics intra‐articular injections at the two‐ or three‐month time points (mean ± SD VAS −33.9 ± 23.9; mean ± SD WOMAC −17.2 ± 3.9) and (mean ± SD VAS −22.9 ± 10.5; mean ± SD WOMAC −16.62 ± 6.2). Nonbiologics placebo WOMAC pain reduction increased between one and four months (mean ± SD WOMAC −3.4 ± 1.0 and −25.6 ± 10.8; P = 0.047). By six months, nonbiologics trended toward a greater mean reduction in pain than biologics (mean −17.0, 95% CI −9.7 to −24.5 vs mean −6.3, 95% CI −3.9 to −8.7). Only one subcategory, naturopathic topical cream, failed to produce a reduction in pain greater than the MCID at any of the time points analyzed. No oral placebo subcategory had a reduction in pain greater than any intra‐articular placebo subcategory (Figures 2 and 3, Table 3). Detailed meta‐analyses per time point are shown in Supplementary Figure 1.

Figure 3.

Placebo effect for VAS pain categorized by subcategory of active treatment. Data are mean and 95% confidence intervals from individual meta‐analyses per time point (Supplementary Figure 1). MCID, minimally clinically important difference; VAS, visual analog scale.

Table 3.

Adjusted P values for differences in placebo pain reduction between subcategories of active treatment^*

	Months	Nonbiologic		Medication		Supplement		Synthetic		Naturopathic
		WOMAC	VAS	WOMAC	VAS	WOMAC	VAS	WOMAC	VAS	WOMAC	VAS
Biologic	1	<0.01.	0.53	<0.01	<0.01	<0.01	<0.01	0.99	0.28	0.02	0.18
	1.5	0.97	>0.99	0.79	0.99	0.61	0.88	>0.99	>0.99	0.92
	2	0.12	0.66	0.15	0.01	0.08	<0.01		0.20
	3	>0.99	0.98	0.17	0.48	0.06	0.12	0.79	0.95
	5			0.07		0.20
	6	0.30	0.88	0.80	0.94	0.37
	12		>0.99		0.81
Nonbiologic	1			>0.99	<0.01	0.48	<0.01	<0.01	0.88	0.36	0.62
	1.5			0.27	>0.99	0.19	0.53	0.95	0.96	0.68
	2			0.82	<0.01	0.96	<0.01		0.50
	3			0.14	0.01	<0.05	<0.01	0.76	1.00
	4			0.03	0.55	0.13	0.74
	6			0.57	0.42	>0.99
	12				0.86
Medication	1					0.60	0.75	<0.01	0.11	0.46	0.97
	1.5					>0.99	0.38	0.91	0.96	>.99
	2					0.96	0.90		0.93
	3					0.88	0.55	<0.01	0.09
	4					0.77	0.96
	5					0.19
	6					0.67
Supplement	1							<0.01	0.02	0.99	0.75
	1.5							0.76	0.98	>0.99
	2								0.73
	3							<0.01	0.01
Synthetic	1									0.07	0.96
	1.5									0.96

^*Bold values are adjusted P‐values below 0.05. VAS, visual analog scale; WOMAC, Western Ontario and McMaster Universities Arthritis Index.

DISCUSSION

As far as we are aware, this is the largest, most comprehensive meta‐analysis of the placebo effect in KOA providing evaluation of more RCT evidence to help quantify the profound effect on pain. Our meta‐analysis of 129 RCTs demonstrates the pain reduction associated with the placebo effect varies not only with category, but also subcategory of active treatment. Furthermore, we found heterogenous changes in pain reduction over time between different placebo modalities.

With more than 90% subcategory data points having a significant reduction in pain from baseline and almost half having a pain reduction greater than MCID, these results are compelling evidence supporting the idea that the placebo effect must not be overlooked when evaluating KOA therapies. Our findings are consistent with the profound and clinically relevant placebo response previously reported and also align with the idea that intra‐articular placebos have a larger effect on pain reduction than oral placebos. ²² , ²³ No previous study has looked for variation in the placebo effect for placebos associated with different subcategories of active treatment within the same modality. We found that intra‐articular saline injections associated with biologics have a significantly greater reduction in pain than intra‐articular saline injections associated with nonbiologics at one month, suggesting the subcategory of active treatment may contribute to the power of the placebo.

This review has major implications for the findings of meta‐analyses comparing multiple subcategories of active treatments for KOA. The pain reduction in placebo treatment is influenced by the active treatment, so comparing treatment categories by subtracting out respective placebo effects may underestimate the potential therapeutic benefit of certain treatments. This concept has previously been discussed in context of the placebo responses between different categories of KOA treatment. Zhang ²² gave the example that oral medication may perform significantly better than oral placebo, whereas intra‐articular active treatment may not perform better than intra‐articular placebo; however, both intra‐articular active treatment and placebo may perform better than oral medication. In this case, intra‐articular treatment should be preferred to oral treatment even though oral medication outperformed its placebo by a greater margin. The idea that differential placebo effects can alter the relative ranking of therapies was further supported by Bannuru et al. ²³ This study found that, although NSAID oral treatments were better than their corresponding placebo and hyaluronic acid intra‐articular treatments were not, the hyaluronic acid intra‐articular treatments were in fact more effective than NSAID oral treatments when compared directly. Our study suggests the same theory can be applied to subcategories of intra‐articular injections. For example, at one month, a biologic intra‐articular injection may not improve pain over its placebo injection, but it may still be more effective at reducing pain than a nonbiologic treatment that is superior to its placebo. There has been discussion of implementing sham injections or no‐treatment arms in future RCTs to further understand the efficacy of intra‐articular saline. ⁶ , ⁷ Designing and conducting these studies to resolve actual therapeutic and contextual effects is critical to help elucidate the effects of the category of treatment, the role of the active treatment, and the potential intrinsic benefits of saline itself.

The difference in pain reduction did not reach statistical significance within the subcategories of oral or topical treatments. This may be a result of less variation in the factors that affect the placebo response between these treatments. Placebos associated with intra‐articular biologic therapies have been described as an “ideal placebo” due to their branding, cost, high expectations, novelty, and route of delivery. ⁸ Although route of delivery is the same between biologic and nonbiologic treatments, all the other factors favor an increased placebo response in biologic intra‐articular injections. Centers offering biologic injections claim about 80% of patients had “good results” or “symptomatic improvement,” above what is published in literature reports. These elevated claims along with the cost, usually approximately $5,000 out‐of‐pocket, lead to increased expectations. ⁵ , ²⁴ Suarez‐Almazor et al ²⁵ found patients with higher expectations experienced a greater pain reduction in KOA than patients who did not. Our results support the idea that treatment expectations influence the placebo effects, and that equivalent treatment administration does not sufficiently account for variation in the placebo response within intra‐articular injections.

We also found the placebo effect for different treatment modalities varies over time. The WOMAC pain reduction for placebo biologic intra‐articular injections decreases by more than 50% between two and six months, whereas the WOMAC pain reduction for synthetic placebo treatments increases by more than 50% between two and six months. At one month, identical intra‐articular placebo methodologies led to different amounts of pain reduction based on the subcategory of active treatment they were associated with (P < 0.0001). This relationship reverses at 6 month and 12 months, with nonbiologics trending toward a greater mean pain reduction than biologics. Oral placebo studies mirrored the intra‐articular changes but were less profound and did not reach statistical significance. Despite both groups receiving inert pills, placebos given in medication trials led to a greater mean decrease in pain than placebos given in supplement trials at early time points. This trend reverses at later time points. These changes raise questions around the accuracy of combining results spanning multiple months, particularly when comparing different placebo modalities. The variations in placebo effect over time should be considered when designing and interpreting studies.

Although the primary analysis using the random effects model revealed group effects in this meta‐analysis, a secondary analysis would ideally be performed to discern between statistical and clinical effects. The goal of such a secondary analysis is to discern the magnitude of the heterogeneity between studies affecting the effect size. For example, effects may be driven by different study protocols (eg, dose and frequency of treatment), the investigated cohort (eg, age, obesity, disease severity and chronicity, comorbidities). Despite our broad search criteria and large number of studies included, this review is limited by the number of studies per time point. For this reason, it was impossible to incorporate heterogeneity into the estimate of the placebo effect its effect size should be interpreted carefully. Although not ideal, this is in line with the Cochrane Handbook for Systematic Reviews of Interventions, which suggests that when data are sparse, the primary concern is to discern whether there is any signal of an effect. ²⁰ Binning results into longer time intervals would have increased the power of the data by increasing the number of studies associated with each point in time. However, we decided not to bin the data to maintain granularity, given that little is understood about the influence that duration of follow‐up has on the placebo effect in KOA. We grouped articles based on methodology of active treatment; however, studies used different protocols with a wide range of dose and administration schedules. A previous meta‐analysis showed no influence of procedure‐related variables, such as the amount of solution injected or the number of injections performed ²⁶ ; however, we cannot exclude the possibility that variation in protocols may play a role in the placebo effect (as mentioned previously). The often‐fluctuating course of symptoms in KOA further complicates quantification of the placebo effect.

Another limitation of the study was the wide use of rescue medication. More than three‐quarters of the studies included allowed for the use of an oral analgesic, typically acetaminophen, as a rescue medication. Twenty percent of articles did not report a specific limit, and 24% did not report on the use of rescue medication, making it challenging to further investigate the impact that rescue medication has on KOA trials. Creating a standard universal rescue analgesic protocol including reporting requirements may be necessary to fully understand the placebo effect on KOA treatment. This analysis focused on treatments of KOA, and it is possible that different magnitudes of placebo effects may be detected in different joints or in different pathologies. Furthermore, P values adjusted for multiple comparisons were used to report significances when comparing the different treatments per time point, which led to nonsignificant comparisons on some occasions even when 95% CIs did not overlap (Figures 2 and 3). When comparing two individual data points in Figures 2 and 3, nonoverlapping 95% CIs indicate significance only when other data are excluded from the statistical analysis (ie, an individual t‐test is performed). Another limitation was that studies were included that (1) appeared not to be double‐masked and (2) had increased risk of bias in the treatment group due to industry funding (Supplementary Table 3). At this moment, it is unclear from the current study by which magnitude this may or may not affect the placebo effect. Furthermore, it is possible that different treatment regimens, doses, and primary end points per subgroup influence the placebo effect. The number of available studies did not allow for such a granular investigation, but it may be investigated in the future if either more studies are published, or treatment regimens and doses become more standardized.

Effectiveness of a treatment clinically does not equate to improvement over a placebo. However, current research on intra‐articular treatments still uses improvement over placebo to compare subcategories. ²⁷ This has been the standard for decades, using the “placebo effect” to describe the underexamined factors that are difficult to quantify. However, if the method alone is evaluated without considering the overall patient outcome, it implies that successful treatment of a patient is limited to the aspects of medical intervention we have scientifically quantified. This paradigm devalues the power of the placebo and the many factors that contribute to it. Although this study shows that understanding the confounding factors behind placebo response is critical both in assessing the efficacy of novel KOA therapies and in comparing the effectiveness of historical treatment methodologies, it has much wider reaching implications. Whether we choose to harness this profound effect or merely take it into account when studying novel treatments, it is crucial that we continue to study and delineate how different subcategories of active treatments lead to varying placebo effects.

AUTHOR CONTRIBUTIONS

All authors contributed to at least one of the following manuscript preparation roles: conceptualization AND/OR methodology, software, investigation, formal analysis, data curation, visualization, and validation AND drafting or reviewing/editing the final draft. As corresponding author, Dr Ward confirms that all authors have provided the final approval of the version to be published, and takes responsibility for the affirmations regarding article submission (eg, not under consideration by another journal), the integrity of the data presented, and the statements regarding compliance with institutional review board/Declaration of Helsinki requirements.

Supporting information

Disclosure form.

Appendix Table 1: Search Criteria

Appendix Table 2: Included Papers

Appendix 3: Bias Assessment

Appendix 4: Individual meta‐analyses of placebo effects