Bright light therapy for mental and behavioral illness: A systematic umbrella review

ABSTRACT Bright light therapy (BLT) is a promising non-pharmacological treatment for a range of psychiatric conditions. The goal of this review was to provide a comprehensive overview of the efficacy of BLT across mental and behavioral illnesses. Using systematic umbrella review methodology, we searched Ovid MEDLINE, Embase, PsycInfo, Web of Science, and Google Scholar for systematic reviews of randomized controlled trials (RCTs) evaluating BLT for any mental or behavioral illness from the date of inception until March 2021. Review quality was assessed using the AMSTAR 2 tool and summary efficacy data were extracted from recent reviews. Of 792 unique records, 67 systematic reviews were included which targeted a range of disorders related to mood, neurocognition, sleep, and eating. Recent meta-analyses targeting seasonal or non-seasonal depression found that BLT outperformed light-related control conditions. Reviews of other disorders identified few RCTs and generally did not support the efficacy of BLT for various outcomes. Overall, the extant literature supports the efficacy of BLT for seasonal and non-seasonal depression, although higher quality systematic reviews are needed to increase confidence in these findings. There was no specific funding for this review, and it was preregistered on Prospero (ID: CRD42021240751).


Introduction
Nearly 40 years ago, Rosenthal et al. (1984) found that ocular exposure to experimental light reduced symptoms of seasonal depression. Since then, bright light therapy (BLT), or the therapeutic application of visible light, has seen increasing interest within psychiatric research and practice, and has been investigated as a potential treatment for various mental and behavioral illnesses, including mood disorders, eating disorders, attention-deficit/hyperactivity disorder, and posttraumatic stress disorder (Campbell et al. 2017;Zalta et al. 2019). Part of this popularity comes from an expanded awareness of the different neurobiological mechanisms that can give rise to mental illness, one of which, implicated by emerging evidence in a range of diagnostic conditions, is dysfunction within the circadian system (Carpenter et al. 2021;McGowan and Saunders 2021;Meyer et al. 2020). Given that visual light is the primary stimulus by which the central circadian clock is synchronized with the external environment, BLT is believed to provide beneficial effects by correcting disturbances within circadian rhythms (Oldham and Ciraulo 2014a). BLT is also thought to act upon several other pathways broadly implicated in mental health, including monoaminergic signaling (e.g., serotonin, melatonin) and autonomic functioning (Campbell et al. 2017;Oldham and Ciraulo 2014a). Such findings suggest that BLT may have a broad mental health impact not confined to a specific diagnostic condition. As the efficacy of traditional pharmacological and psychotherapeutic treatments tends to vary across mental and behavioral illnesses, BLT stands out as a promising approach that may help reduce the considerable global burden inflicted by such disorders (Leichsenring et al. 2022;Vigo et al. 2016).
The focus on BLT has resulted in a growing number of published trials, which, in turn, has led to a proliferation of review papers ( Figure S1). Not all reviews are equal, however, and systematic reviews (SRs) of randomized controlled trials (RCTs) provide the highest level of evidence regarding the efficacy of an intervention (Murad et al. 2016). The sheer number of BLT-related reviews (not to mention the overlap in search terms with non-ocular forms of light therapy) may make it difficult for health professionals to "strike gold" and identify the SRs that allow them to best assess the scope of the evidence surrounding BLT. A systematic umbrella review, or an SR of SRs, is a powerful method of providing a comprehensive, high-level overview of a research topic (Higgins et al. 2022). Importantly, umbrella reviews can provide a user-friendly form of research synthesis, allowing readers to easily digest a broad range of evidence, such as the efficacy of an intervention across numerous disorders (Higgins et al. 2022). Accordingly, the goal of this paper was to comprehensively summarize the current state of evidence regarding the efficacy of BLT for mental and behavioral illness using umbrella review methodology.

Search strategy and selection process
The following databases were searched from the date of inception until March 2021: Ovid MEDLINE, Embase, Web of Science Core Collection, PsycINFO, and Google Scholar. Key words and MeSH terms focused on light therapy, mental or behavioral health, and SRs or metaanalyses (see Table S1 for full search strategies). Titles and abstracts of results were screened to identify potential SRs, followed by full-text evaluation to determine inclusion status. Citation searching of relevant references from identified SRs were also examined. Screening and full text eligibility evaluation were both performed by two independent reviewers, with disagreements settled by a third independent reviewer, and final inclusion status of each record was discussed until consensus was reached. The flow chart for study selection can be seen in Figure 1. A list of articles excluded at full text review with associated justifications are available in Table S2.

Review eligibility
Reviews were eligible for inclusion based on the PICOTS framework: 1) Population -targeted human patients with relatively homogeneous presentations of mental or behavioral health disorders; 2) Intervention -reviewed RCTs of BLT; 3) Control/ comparator group -included RCTs that employed a light-related control; 4) Outcome -examined the efficacy of BLT; 5) Timing -not applicable, there were no restrictions on timing of the intervention or outcome assessment; and 6) Study design -were an SR with or without a meta-analysis (Page et al. 2021).
Regarding the study design, SRs were defined as reviews having: a systematic, reproducible search strategy and a description of the process by which studies were included/excluded; a comprehensive search using a minimum of two scientific databases; a description or list of included studies; and an evaluation of the risk of bias (RoB) or quality of each included study using a standardized process. Conference abstracts, articles without an English translation, and inaccessible articles were also excluded. Individual studies included in each potential SR were examined to confirm that at least one was an RCT that compared BLT -broadly defined as any retinal exposure to a visible light-emitting device intended to have a therapeutic effect -against a light-related control. Due to the lack of consensus as to what constitutes an appropriate control for BLT, light-related controls were broadly defined as any condition that employed one of the following: dim light, blue-depleted light (e.g., red, amber), or, due to their traditional use as a BLT comparator, negative ion generators. However, control conditions were required to match the procedure for the active treatment. For example, if an RCT used a light box for treatment, control participants needed to be exposed to a light-related control box as well. Additionally, all adjunctive treatments, including pharmaceutical or behavioral interventions, needed to be identical across groups, so that the only difference between the treatment and comparator groups was that the former received a light intervention and the latter received a light-related control. This operationalization of lightrelated controls was chosen to ensure that the extant BLT literature was thoroughly examined, while also, as much as possible, isolating the therapeutic effects of light and mitigating the impact of non-specific treatment factors. Finally, studies that provided incomplete descriptions of their group assignment or crossover allocation process were considered randomized only if they provided sufficient detail that made it likely that randomization had indeed occurred (e.g., statistical predetermination of group assignment, balanced crossover).
Reviews were excluded if they used a bias/quality assessment that was incapable of distinguishing between individual RCTs. Examples of this include "levels of evidence" approaches that rated quality solely on overall study design (e.g., rating all RCTs as equal strength) and global evaluations that pertained only to the collective body of included studies without reporting individual RCT RoB.

Data collection and quality assessment
All data were extracted by one reviewer and confirmed by a second. Data were extracted from available information using a custom data form and included the following: 1) SR characteristics (e.g., number of RCTs, RoB/study quality tool); and 2) SR outcomes. SR quality was assessed using the AMSTAR 2 tool, which appraises quality of SRs of healthcare interventions by evaluating critical and non-critical methodological domains (Shea et al. 2017). The AMSTAR 2 tool provides an overall rating for the confidence in reported results as either "high," "moderate," "low," or "very low." SR quality was rated by two independent reviewers and disagreements resolved via consensus discussion.

Data synthesis
To limit the degree of overlap for SRs on the same disorder, interpretation focused on the most recent SR for a given outcome within a disorder, in line with recommendations by Higgins et al. (2022). Metaanalyses were prioritized and selected over more recent SRs if published within the preceding few years. For each highlighted SR, available key findings -including aggregate effect size -were presented. For consistency, directions of standardized mean differences (SMDs) were reversed when needed, so that negative scores equaled a greater decrease in the outcome measure in the BLT group relative to the control group. As such, interpretation of SMDs is dependent upon the directionality of the outcome in question (i.e., which direction is "good"), with negative scores favoring BLT in some cases (e.g., depressive symptom severity) and favoring the control group in others (e.g., total sleep time). This was done to maintain the direction of the original scale for each outcome measure. When the highlighted SRs contained non-RCTs or RCTs that did not use a light-related control, interpretation was made based on only the relevant bright light RCTs. For meta-analyses, this entailed reanalyzing outcome data with the smaller set of studies. To do so, effects size estimates (e.g., SMDs, risk ratios) and variances were extracted from the original SR article and re-analyzed using random-effects meta-analysis in R 4.0.5 (R Core Team 2021), with weighting done by inverse variance.

Results
Of the 792 unique records assessed for eligibility, 67 SRs of bright light RCTs met criteria and were included in the present umbrella review (Figure 1). Details for each SR can be found in Table S3. In total, 26 SRs targeted mood disorders, 33 targeted neurocognitive disorders, six targeted sleep-related disorders, one targeted bulimia nervosa, and one SR included multiple disorders. Figure 2 shows the cumulative number of SRs for each disorder domain over time.
Nearly every SR was rated as having a critical weakness in one or two quality domains via the AMSTAR 2 tool. Based on the AMSTAR 2 rating criteria, 3 (4.5%) SRs were rated at high confidence in results, 15 (22.4%) SRs were rated at low confidence, and 49 (73.1%) SRs were rated at very low confidence (Shea et al. 2017). The two most common critical weaknesses were not providing a list of excluded studies with exclusion justification (88% of SRs), followed by a lack of review preregistration (63% of SRs). Because the requirement to list excluded studies was only recently added to the PRISMA 2020 guidelines, alternative post-hoc quality ratings were calculated treating this weakness as noncritical, which resulted in 4 (6%) SRs rated at high confidence, 10 (14.9%) SRs rated at moderate confidence, 43 (64.2%) SRs rated at low confidence, and 10 (14.9%) SRs rated at very low confidence (Page et al. 2021). AMSTAR 2 quality ratings for each included SR can be found in Table S4.

Mood disorders
Key findings from recent SRs targeting mood disorders are presented in Table 1. Pjrek et al. (2020) reviewed 19 RCTs examining BLT in participants diagnosed with unipolar or bipolar seasonal depression. In their metaanalysis, the authors found that BLT was associated with a small-to-medium decrease in depressive symptom severity and a better response rate compared to controls. Tao et al. (2020) reviewed RCTs that evaluated the efficacy of BLT in participants with non-seasonal depression, defined as major depressive disorder, persistent depressive disorder, bipolar disorder with depression, or clinically significant non-seasonal depressive symptoms. This meta-analysis included 19 RCTs with 20 unique sample comparisons that compared BLT against a light-related control. Re-analysis of study results, including only RCTs with a light-related control, revealed that BLT was associated with a small-tomedium decrease in depressive symptom severity compared to controls. Sub-group analyses revealed that a similar magnitude of effect size was seen with BLT as a monotherapy (SMD = −0.33, 95% CI: −0.54 -−0.13, p = .004, N = 771, k = 15, I 2 = 33%) and as an adjunctive treatment (SMD = −0.44, 95% CI: −1.09-0.22, p = .139, N = 182, k = 5, I 2 = 46%), although the results for adjunctive BLT did not achieve statistical significance. Lam et al. (2020) meta-analyzed seven RCTs evaluating BLT efficacy in participants who met diagnostic criteria for bipolar disorder (type I or II) and were experiencing a depressive episode. The authors found that BLT was associated with a small-to-medium reduction in depressive symptom severity and an improvement in treatment response odds. Odds of remission, however, did not achieve statistical significance. Additionally, Lam et al. (2020) did not find evidence that BLT was associated with increased odds of affective switching to mania or hypomania episodes. Six of the seven included RCTs did not observe affective switching in either the BLT or control group and overall affective switching rates were below 3%. Smith et al. (2019) reviewed two RCTs examining the impact of BLT on depressive symptoms during pregnancy. Although they were unable to meta-analyze data from the two studies, the authors concluded that BLT was associated with reduced depressive symptom severity in one study of 27 women. However, these results did not reach statistical significance and the other study was noted to have found no difference in depressive scores between groups. Finally, Krasnik et al. (2005)  reviewed studies looking at BLT in participants with premenstrual dysphoric disorder. Meta-analysis of the three RCTs with light-related controls did not find a statistically significant effect of BLT on depressive symptom severity.

Neurocognitive disorders
Key findings from recent SRs targeting neurocognitive disorders are presented in Table 2. Hjetland et al. (2020) reviewed studies broadly examining the effects of BLT on various outcomes in participants with dementia, although findings were not meta-analyzed. Overall, the authors identified 13 papers containing data from nine unique RCTs with light-related controls. Mixed or nonsignificant results were found for the effect of BLT on the following outcomes: depressive symptom severity, behavioral and psychological symptoms of dementia, Mini-Mental State Examination, and activities of daily living. Leng et al. (2020) performed a network meta-analysis comparing the efficacy of different non-pharmacological interventions for agitation in participants with dementia. Four RCTs with five unique comparisons were identified that compared BLT directly against a light-related control. BLT was associated with a small but statistically non-significant increase in agitation. Faulkner et al. (2019) broadly reviewed studies examining the efficacy of BLT on sleep and circadian outcomes in a range of sleep and neuro-psychiatric disorders. Overall, they identified nine RCTs that enrolled participants with dementia. Based on re-analysis of only RCTs with lightrelated controls, BLT was associated with a small decrease in rest-activity rhythm amplitude (i.e., a less vigorous increase in activity when awake versus when sleeping), although it was not associated with changes in other sleep or circadian outcomes. Srisurapanont et al. (2021) performed a network metaanalysis on studies evaluating the efficacy of bluewavelength light therapy (BWLT) for various symptoms following traumatic brain injury. Three of the four included RCTs compared BWLT against an amber light control. Direct comparisons between these two conditions revealed that BWLT was associated with medium or large improvements in sleepiness, depressive symptom severity, and fatigue. No improvement was seen for sleep disturbance. Bannon et al. (2019) meta-analyzed four RCTs evaluating the efficacy of BLT in reducing risk or duration of delirium in critically ill patients, defined as patients in any form of intensive care or high dependence specialty unit. BLT was not associated with a statistically significant reduction in delirium risk. Only one RCT evaluated delirium duration and BLT was associated with an average reduction of delirium length by 0.34 hours, but this difference was also statistically non-significant. Alt. = Alternative AMSTAR 2 rating after scoring Item #7 as a non-critical weakness; BLT = Bright light therapy; ES = Effect size; H = High risk; II = Moderate confidence; III = Low confidence; IV = Very low confidence; I 2 = Meta-analytic heterogeneity; k = Number of unique comparisons (can be greater than # of RCTs); L = Low risk; N = Sample size; NR = Not reported; OR = Odds ratio; RCT = Randomized Controlled Trial; RR = Risk ratio; S = Some risk; SMD = Standardized mean difference; U = Unclear risk.
a RoB/quality ratings were extracted from the original review; b Results were re-analyzed to include only RCTs with light-related controls.

Sleep disorders
Key findings from recent SRs targeting sleep disorders are presented in Table 3. Baglioni et al. (2020) performed a network meta-analysis of various complementary and alternative interventions for insomnia that were given as monotherapies. Two of the included studies compared BLT against a light-related control and BLT was not associated with an improvement in any sleeprelated outcome. As part of their broader review of BLT, Faulkner et al. (2019) included RCTs targeting advanced sleep-wake phase disorder (ASWPD) and delayed sleepwake phase disorder (DSWPD). Two ASWPD RCTs and six DSWPD RCTs used a light-related control. Reanalysis of these RCTs showed that BLT was not associated with improvements in sleep timing or sleep efficiency in participants with ASWPD, or the following outcomes in participants with DSWPD: sleep timing, self-reported daytime sleepiness, sleep onset latency, self-reported sleep quality, self-reported sleep disturbance, or total sleep time. Several SRs focused on populations susceptible to disrupted sleep due to exposure to environmental factors. Miller et al. (2019) reviewed studies of nonpharmacological sleep interventions for patients in inpatient or residential settings, although they did not conduct meta-analyses. Three RCTs evaluated BLT and made use of a light-related control. BLT was not associated with improvements in sleep outcomes, measured either by actigraphy or sleep questionnaires, or melatonin expression. Slanger et al. (2016) performed a Cochrane review of RCTs evaluating non-pharmacological interventions for sleepiness caused by shift work. Four RCTs evaluated BLT and used a light-related control. A meta-analysis of two of the RCTs found that BLT was associated with a large reduction in sleepiness as measured by the Stanford Sleepiness Scale. Fogarty et al. (2016) reviewed studies of complementary and alternative therapies for the treatment of eating disorders. They identified two RCTs that evaluated BLT in participants with bulimia nervosa. While no meta-analyses were conducted, these RCTs found that  a RoB/quality ratings were extracted from the original review; b Results were re-analyzed to include only RCTs with light-related controls; c One crossover study dropped to avoid a unit of analysis error; d Authors presented results from a network meta-analysis; e 99% CI.

Bulimia nervosa
BLT was associated with reduced depressive symptom severity, although neither found an improvement in eating psychopathology. Finally, the two RCTs arrived at mixed results regarding the effect of BLT on the number of binge and/or purge episodes.

Discussion
Across existing SRs, BLT received the strongest support for reducing the severity of depressive symptoms in seasonal and non-seasonal presentations of depressive illness. Recent meta-analyses of BLT for depression have found small-to-medium effect sizes, with evidence that BLT is efficacious as a monotherapy and may have added benefit as an adjunctive treatment to pharmacotherapy. Additionally, each of these meta-analyses found low-to-moderate heterogeneity, supporting the generalizability of these findings beyond the included RCTs. Also, while depression-focused SRs tended to include participants with both unipolar and bipolar depression, meta-analyses specifically targeting the smaller subset of bipolar depression RCTs found that BLT was also efficacious for this patient population, with limited risk of affective switching as a side effect. On the other hand, BLT was not supported as an efficacious treatment for antepartum depression and premenstrual dysphoric disorder; granted, the non-significant findings may have been influenced by the small number of RCTs identified by SRs of these disorders. Overall, the literature base supports the use of BLT for unipolar and bipolar depression, regardless of seasonality. Past research has highlighted the potential to increase BLT utilization, particularly for non-seasonal depression, which, despite an equivalent level of empirical support, is less likely than seasonal depression to be considered by providers as an appropriate indication  for BLT (Fischer et al. 2012;Oldham and Ciraulo 2014b;Winkler-Pjrek et al. 2016). The observed effect sizes for depression were comparable in magnitude to those seen with both antidepressants and psychotherapy, indicating that BLT is a viable first-line treatment for depressive illness (Cuijpers et al. 2019;Munkholm et al. 2019). The accessible, low-risk nature of BLT could enhance patient-centric care by providing an alternative to psycho-and pharmacotherapy for patients who prefer other treatment modalities, thereby reducing the likelihood of treatment dropout and supporting the therapeutic alliance (Windle et al. 2020). BLT also aligns with recent calls for innovation within the field of depression treatment, given that, while existing treatments demonstrate statistically significant improvements over control conditions, they fail to elicit clinically meaningful responses (Leichsenring et al. 2022;Munkholm et al. 2019). This may be due to the large degree of diagnostic heterogeneity present within depression, which can arise from distinct pathophysiological pathways; accordingly, one proposal to enhance the efficacy of depression treatments is to match treatment to specific clinical presentations (Buch and Liston 2021). Although current evidence does not suggest that BLT outperforms standard treatment approaches for depression, emerging research has implicated disturbance of the circadian system in the pathogenesis and maintenance of certain mood disorder cases (Carpenter et al. 2021). If "circadian depression" is indeed a mood disorder phenotype and can be readily detected in clinical practice, then circadian-targeted treatments like BLT may be especially beneficial for this subgroup of patients (Carpenter et al. 2021). As such, BLT may be a powerful component of precision medicine for depression. The cumulative strength of evidence supporting the use of BLT for depression is tempered by the low quality of existing SRs and the high RoB found in many of the individual RCTs. These methodological limitations, across both the SRs and the underlying trials, reduce confidence in the reported results. Importantly, however, the few depression-focused SRs that received higher quality ratings, like Al-Karawi and Jubair (2016) or Perera et al. (2016) found effect sizes of similar or greater magnitude compared to more recent SRs, suggesting that the overall results of recent SRs may not have been adversely affected by the lower quality of methods. Regardless, the findings of the present umbrella review emphasize the urgent need for enhanced methodological rigor not only in the individual trials but also in the SRs synthesizing the literature. Authors should take care to follow established guidelines for SRs and meta-analyses, both in the conduct and reporting of their review, to reduce the possibility of bias (Higgins et al. 2022;Page et al. 2021).
Unexpectedly, over half of the SRs targeted disorders other than depression. Dementia was most commonly reviewed, followed by brain injuries, delirium, and sleep disorders. Often, RCTs evaluating BLT were captured as part of a broader SR of non-pharmacological interventions for a given disorder. Despite their frequency, nonmood disorder SRs included very few RCTs, meaning that their findings were based on synthesis of limited data. The small pool of available evidence makes it difficult to determine if the generally non-significant findings reported by these SRs are reflective of a lack of BLT efficacy for a given disorder/outcome combination, or if they can be attributed to insufficient meta-analytic power (Jackson and Turner 2017). Promising lines of research identified by this umbrella review that are currently underdeveloped include the use of BLT to treat symptoms associated with traumatic brain injury, alleviate sleepiness in shift workers, and reduce depressive symptoms in patients with bulimia nervosa (Fogarty et al. 2016;Slanger et al. 2016;Srisurapanont et al. 2021). Also, given the primary role of the circadian system, and therefore light, in sleep processes, BLT certainly warrants additional investigation as a potential treatment for sleep disorders (Borbély et al. 2016). The limited number of RCTs is particularly concerning for circadian rhythm sleep-wake disorders, as BLT is recommended as a first-line treatment despite an overall absence of critical empirical support (Duffy et al. 2021;Faulkner et al. 2019). These findings reveal that the full extent of BLT's clinical utility has yet to be adequately investigated.
Most SRs included few details on how they defined BLT or control treatments, especially when BLT was included as part of a broader search for nonpharmacological interventions. Similarly, the characteristics of BLT (e.g., light intensity, timing, frequency, duration) varied considerably across the individual RCTs that were captured in the identified SRs, making it difficult to provide specific recommendations. RCTs targeting neurocognitive disorders primarily used light boxes, although there was little consistency across BLT parameters such as intensity, duration, or timing, which may help explain the overall lack of efficacy seen with these conditions. When BLT was defined in the SR methods, the operationalization was typically vague. For example, Hjetland et al. (2020) defined BLT as an "enhanced indoor electrical light scheme" aimed at altering psychological and behavioral responses. A few RCTs taking place in inpatient, residential, or occupational settings employed the intervention through altered overhead lighting (Bannon et al. 2019;Hjetland et al. 2020;Slanger et al. 2016). The most consistent BLT methods were seen in RCTs targeting depressive disorders, where BLT was generally administered via a light box for 30-60 minutes each morning at an intensity of 2,500-10,000 lux (Pjrek et al. 2020;Tao et al. 2020). Such characteristics are broadly consistent with previous guidelines for administering BLT for the treatment of mood disorders and, at the present, have the strongest support for efficacy (Pail et al. 2011). However, additional research is needed to optimize the administration of BLT, such as light intensity and duration, as well as determine how BLT might be altered to better align with individual patient needs. For example, the effect of light exposure on the circadian system is dependent on the time of administration relative to the patient's biological day/night cycle; as such, BLT may have greater therapeutic effects if combined with objective measurement of circadian timing, such as dim light melatonin onset (St Hilaire et al. 2012).
One limitation of the present umbrella review is that it did not perform a direct search for individual RCTs. Even though the intention was to provide a high-level overview of existing BLT RCTs using recent SRs, it is likely that the included SRs failed to capture all relevant studies, particularly in emerging areas where there may not be sufficient research to justify a full SR. Also, any RCTs published within the few years prior to the end of the literature search date were likely to have been missed, given the length of time needed to both complete and update SRs (Bashir et al. 2018;Borah et al. 2017). Another limitation concerned the grouping of the specific disorders. SRs varied considerably in the criteria they used to define disorders of interest, particularly when targeting mood disorders. For example, various forms of depression -such as major depressive disorder, bipolar disorder (I or II), and persistent depressive disorder -were often combined for data synthesis, which may have resulted in an incomplete picture regarding the efficacy of BLT. While such methodological decisions did not lead to high meta-analytic heterogeneity in these SRs, it is still possible that BLT does not have equal efficacy across different forms of depressive illness. Finally, this review was limited to SRs written in English, meaning that important contributions to the literature may have been missed.
In conclusion, the extant literature supports the efficacy of BLT for seasonal and non-seasonal depression, although higher quality SRs are needed to increase confidence in these findings. SRs of BLT for non-mood disorders included few RCTs, limiting the inferences that can be drawn from the reported results.

Disclosure statement
No potential conflict of interest was reported by the author(s).

Funding
There was no specific funding for this review.

Author contributions
DR and NB contributed to review conceptualization. DR contributed to data visualization. All authors contributed to study design, data collection, data interpretation, and drafting of the manuscript.