Placebo administration for dry eye disease: a level I evidence based systematic review and meta-analysis

Background The efficacy of various common treatment options for dry eye disease (DED) has been investigated against placebo. However, the potential beneficial effect of placebo in the management of DED is still unclear. Aim This meta-analysis investigated the impact of placebo administration in DED in Ocular Surface Disease Index (OSDI), Schirmer I test (SIT), tear breakup time (TBUT), corneal staining, and complications. Method This meta-analysis and systematic review was conducted according to the 2020 PRISMA guidelines. In March 2022, Pubmed, Web of Science, Google Scholar, and Embase were accessed. All the randomised clinical trials which investigated any active treatment against a placebo control group were considered. The following data were extracted at baseline and at last follow-up: Ocular Surface Disease Index (OSDI), tear breakup time test (TBUT), Schirmer I test (SIT), corneal staining. Results Data from 56 studies (12,205 patients) were retrieved. Placebo administration is not effective in improving TBUT (P = 0.3), OSDI (P = 0.2), SIT (P = 0.1) and corneal staining (P = 0.1) from baseline to last follow-up. Active treatment led to a higher TBUT and SIT compared to placebo administration (P < 0.0001). The active treatment resulted in a lower OSDI compared to placebo administration (P = 0.0005). Five studies reported data on the corneal staining. No difference was found between placebo administration and active treatment (P = 0.8). Conclusion Placebo administration does not impact symptoms of DED and can be successfully employed to evaluate the efficacy of active treatments. Supplementary Information The online version contains supplementary material available at 10.1007/s11096-022-01439-y.


Introduction
Dry eye disease (DED) is a common condition of the ocular surface [1], with a prevalence of up to 50% of the global population [2]. Etiological factors include ocular surface inflammation and damage, neurosensory abnormalities, and tear film instability, which is caused by insufficient tear production or quality of the tear film [3]. The main subtypes of DED are aqueous deficient and evaporative DED, with frequent co-existence of both subtypes [2]. DED results in visual disturbance, burning, pain, and photophobia [4,5]. Conventional therapies for DED include artificial tears, punctal occlusion, topical corticosteroids or secretagogues, and oral essential fatty acid 1 3 supplementation [6][7][8][9]. Artificial tears are a mainstay therapy as they provide an affordable and immediate relief [10]. However, as inflammation is a key component in the pathogenesis of DED, artificial tears might be inadequate in improving the ocular surface damage in patients with more severe DED. Recently, the importance of drugs with anti-inflammatory or secretagogue properties has been highlighted in such cases [11].
In randomized controlled trials (RCTs), the efficacy of an active treatment is commonly evaluated by the difference in outcome between the intervention and placebo group [12]. The placebo effect describes a phenomenon of improvement of symptoms in patients receiving an inert substance [13]. Placebo administration can actually be more beneficial than no-treatment in many clinical settings [14]. Therefore, the clinical impact of a placebo might be neglected [15], resulting in an "efficacy paradox" [16]. This term describes the discrepancy between the treatment efficacy suggested by RCTs and the treatment efficacy observed in the clinical practice [16]. The rate of placebo effect in RCTs investigating various conditions is estimated at 30-40% [17][18][19]. The placebo effect lowers the statistical power of RCTs, challenging the interpretation of the treatment effects [20]. Placebo effects have been attributed to complex processes, such as behavioural conditioning, patients' expectations, regression to the mean, and the Hawthorne effect [21][22][23][24]. Regression to the mean describes that extreme outliers tend to a more average value [24]. In clinical practice, patients with a symptomatic condition often tend to improve spontaneously, even without treatment [24]. The Hawthorne effect relates to a change in behaviour of individuals in response to the perception of being observed [25]. Therefore, it affects the generalisability of RCTs to clinical practice [25]. Placebo administration is a major methodological challenge in RCTs investigating DED [27]. The special feature of DED trials is that placebo administration might involve adding a liquid to the ocular surface with possible inherent therapeutic effects in DED [27]. Recently, the processes involved in inducing a placebo effect have been investigated, but the exact mechanisms underlying the placebo effects are not fully understood yet [26]. In some RCTs, patients received a placebo prior to the randomisation procedure to reduce the placebo effects during the treatment period [9,28,29]. However, placebo effects were evidenced also in these studies [27]. The efficacy of various common treatment options for DED has been investigated in the clinical setting using placebo as a comparator [7,9,.

Aim
This meta-analysis investigated the impact of placebo administration in DED in Ocular Surface Disease Index (OSDI), Schirmer I test (SIT), tear breakup time (TBUT), corneal staining, and complications.

Eligibility criteria
All the randomised clinical trials which investigated any active treatment with a placebo control group were accessed. According to the authors language capabilities, articles in English, German, Italian, French and Spanish were eligible. Only level I of evidence studies, according to Oxford Centre of Evidence-Based Medicine [86], were considered. Only studies published in peer reviewed journals were considered. Reviews, opinions, letters, editorials were not considered. Animal and in vitro studies were not eligible.

Search strategy
This meta-analysis and systematic review was conducted according to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses: the 2020 PRISMA statement [87]. The PICOT algorithm was preliminary pointed out:

Selection and data collection
Two authors (F.M. and J.P.) independently performed the database search. All the resulting titles were screened and, if suitable, the abstract was accessed. The full-texts of the abstracts which matched the topic of interest were accessed. If the full-text of the article was not retrievable or accessible, the study was excluded. The bibliography of the full-text articles was also screened by hand. Any disagreements were discussed and settled by consensus.

Assessment of the risk of bias and quality of recommendations
The between studies risk of bias assessment was evaluated using the risk of bias tool of the Review Manager software (The Nordic Cochrane Collaboration, Copenhagen). The following biases were evaluated by an author independently (J.P.): selection, performance, detection, attrition, reporting, other sources of bias. To investigate the overall risk of publication bias, the funnel plot of the most reported outcome was performed. To grade the quality (or certainty) of evidence and strength of recommendations, the Grading of Recommendations Assessment, Development and Evaluation (GRADE) was performed [91].

Synthesis methods
The statistical analysis was performed by one authors (F.M.). To assess the improvement from baseline to the last follow-up, the IBM SPSS software version 25 was used. Mean difference (MD) and unpaired t-test were evaluated. For the comparisons, a meta-analysis was conducted using the Review Manager software (The Nordic Cochrane Collaboration, Copenhagen) version 5.3. Data were analyzed using the inverse variance and MD effect measure. The comparisons were performed with a fixed model effect as set up. Heterogeneity was assessed through the 2 and the Higgins-I2 test. If 2 < 0.05 and I 2 test > 60%, a random model effect was adopted. The confidence intervals (CI) were set at 95% in all analyses. Values of P < 0.05 were considered statistically significant. Forest plots were performed for each comparison.

Study risk of bias assessment
Given the randomised nature of the studies selected, the risk of selection bias was low. Most authors performed blinding of participants, personnel, and assessor, leading to an overall low risk of performance and detection biases. Equally, the risk of detection and attrition biases were low. Finally, the risk of other bias was low to moderate. Concluding, the risk of bias graph evidenced a good quality of the methodological assessment (Fig. 2).

Risk of publication bias
To assess the overall risk of publication bias, the funnel plot of the most commonly reported outcome was performed and evaluated (SIT). The plot evidenced good symmetry, and the effects were located withing the pyramidal shapes of acceptability (Fig. 3). This indicates a low risk of publication bias.

Study characteristics and results of individual studies
Data from 4934 patients were retrieved. 73% (3602 of 4934 patients) were women. The mean follow-up was 12.2 ± 16.0 weeks. The mean age was 50.3 ± 14.5 years. Baseline comparability between the placebo and active treatment was found in terms of TBUT (P = 0.8), OSDI (P = 0.7), SIT (P = 0.7), corneal staining (P = 0.7). Generalities of the included studies are reported in Table 1, and the analysis of the baseline comparability is shown in Supplementary material 2.

Quality of the recommendations
The level of evidence quality according to the GRADE system was high for TBUT, SIT, and OSDI, whereas a low level of evidence in quality for the corneal staining was evidenced (Fig. 5).

Statement of key findings
The present study demonstrated that placebo administration did not improve symptoms of DED at the last follow-up. Placebo administration can therefore be considered as passive comparator to evaluate the efficacy and safety of active treatments in DED.

Strengths and weaknesses
The literature search was performed in Pubmed, Web of Science, Google Scholar and Embase, as previously recommended to guarantee efficient coverage [115]. Within the selection and data collection, no severe disagreements between the two responsible authors occurred. Therefore, no impact on internal validity must be expected. The present meta-analysis and systematic review was detailed and precise, but has several limitations. Firstly, the active treatment and the placebos protocols showed a high variability. Hypothetically, different effects might be attributable to different placebos. Selek et al. and Goldstein et al. attributed the symptomatic improvement witnessed in the placebo group to a lubricating effect of the vehicle itself on the ocular surface [73]. Shin et al. evaluated the efficacy of acupuncture for DED, and employed a sham acupuncture control group which they referred to as the placebo group [92]. Yet, placebo acupuncture is technically impossible. Accordingly, "sham" describes any control treatment of acupuncture aiming to make the patients believe that they received the real treatment [116,117]. A superior efficacy of sham acupuncture compared to pharmacological placebo has been suggested [118]. Conversely, patients demonstrating a therapeutic effect to placebo were excluded after a 2-week-run-in period. Thereby, the risk of a strong placebo effect should be mitigated [61]. Moreover, the heterogeneous length of follow-up might also limit the reliability of our results. Given the lack of quantitative data available for inclusion, no further subgroups were possible to investigate.

Interpretation
Previously, Imanaka et al. investigated predictive factors of the placebo effect in trials for DED, collecting data from 205 patients enrolled in 3 placebo-controlled RCTs. High baseline scores and age affected the placebo responses of the corneal staining score [27]. Ageing is an essential risk factor for DED [27]. The high proportion of females (73%) in the present study agrees with previous publications and has been attributed to the effects of sex steroids including oestrogens, glucocorticoids, and epigenetics [93]. While the exact association of sex steroids and DED is still unclear, relatively low levels of serum androgen in females are associated to lower anti-inflammatory effects on the ocular surface, promoting DED [94]. Hypothetically, the placebo might have a beneficial effect in patients with high baseline scores by acting as artificial tears [27]. Also, the greater placebo effect in patients with high baseline scores can be attributed to the regression to the mean phenomenon [24].
In other conditions, such as low back pain, a previous metaanalysis showed rapid symptom improvement in the first 6 weeks and less marked improvement thereafter both in the treatment and placebo groups [95]. Therefore, regression to the mean might commonly be falsely interpreted as efficacy, but occurs simply by chance and with time [96]. Placebo effects have been reported in patients with depression, cardiovascular diseases, asthma, and different pain syndromes [97][98][99][100][101]. Notably, a systematic review collecting data from 72 RCTs with 9827 patients with fibromyalgia and 70 RCTs with 10,297 patients with diabetic neuropathy showed a superior effect of placebo administration compared to non-treatment. The placebo group showed a significant reduction of pain and fatigue [102]. It has been presumed that the examination by a doctor and the actual physical motions of taking medications can have therapeutic effects [96]. Placebo effects have been attributed to the release of substances such as endogenous opioids [103], endocannabinoids [104], dopamine [105], oxytocin [106], and vasopressin [107]. These substances show specific effects to the target system [101]. In addition, neuroimaging studies have shown changes similar to those caused by opioids in brain activation patterns induced by placebo [108]. In DED, patients show positive perceptions regarding the effectiveness of their treatments [109]. The improvement anticipated by the patient might be attributable to the measured improvement [110]. Notably, some of the studies included in this meta-analysis showed some beneficial effect of placebo on SIT [81,83], TBUT [31,75], corneal staining [53,66,75] or OSDI [7,75,79]. The small, nonstatistical reduction of corneal staining induced by placebo may well be regression to the mean [53]. Chang et al. investigated the efficacy of oral uridine compared to placebo on DED. They reported a reduction of corneal staining and improvement of SIT in the uridine compared to the placebo group with statistically significant differences between the groups. Moreover, treatment with uridine significantly reduced the OSDI score. However, the placebo group also showed some reduction of the OSDI score. The authors assume that these nonsignificant differences between the groups are attributable to the placebo effect [79]. In 2018, the Dry Eye Assessment and Management (DREAM) trial concluded that omega-3 fatty acid supplements do not provide better outcomes than placebo in the management of DED [75,111]. As to adverse events, placebo proved safe in the studies considered for our meta-analysis. However, Kinoshita et al. reported that eye irritation occurred more frequently in the placebo than in the treatment (rebamipide) group. However, the authors did not specify what the placebo consisted of [9]. In the DREAM trial, the percentage of patients with at least one serious adverse event was 8.1% in the placebo group receiving olive oil. The percentage of patients with at least one nonserious adverse event was similar in the active treatment group and the placebo group (61.9% and 60.8%, respectively) [75].
In the DREAM extension study, one patient in the placebo group was hospitalized for dyspnoea as a serious adverse event [83]. A randomized study published in 2017 reported that patients were more likely to report adverse events when they were aware that they received statin therapy than when they were blinded [112]. Therefore, negative expectations of the patients regarding the treatment with either active or inert substance might cause a negative placebo effect, called the nocebo effect [112]. Nocebo effects have been described in many clinical contexts. Up to 19% of adult and 26% of elderly patients receiving placebos report side effects [113]. They have been partly attributed to the verbal suggesting in the context of informed-consent process [101]: in the clinical setting, expectancies might be affected by the preceding description of the treatment [101]. On a neurobiological level, the nocebo effect has been shown to be mediated by cholecystokinin [114] and to be associated with hyperactivity of the hypothalamic-pituitary-adrenal axis [101]. This increased activity is antagonized by benzodiazepine, which suggests the role of anxiety of patients in nocebo effects [101].

Further research
Altogether, distinguishing whether an observed effect is secondary to placebo, regression to the mean, or simply part of the course of the disease, can be challenging [96]. The exact intensity of the placebo effects in clinical trials on DED remains difficult to determine. In patients with DED, placebo did not show beneficial therapeutic effects, but was safe. However, no subgroup analysis could be performed given the lack of quantitative data available for inclusion. Therefore, further research is warranted to focus on evaluating different types of placebo administration in DED.

Conclusion
Placebo administration does not impact symptoms of DED and can be successfully administered to evaluate the efficacy of active treatments. These conclusions must be interpreted within the limitations of the present study.
Funding Open Access funding enabled and organized by Projekt DEAL. No specific funding was received.

Conflicts of interest
The authors have no conflicts of interest to declare.
Open Access This article is licensed under a Creative Commons Attribution 4.0 International License, which permits use, sharing, adaptation, distribution and reproduction in any medium or format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made. The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder. To view a copy of this licence, visit http:// creat iveco mmons. org/ licen ses/ by/4. 0/.