The impact of rising temperatures on the prevalence of coral diseases and its predictability: a global meta-analysis

: Coral reefs are under threat from disease as climate change alters environmental conditions. Rising temperatures exacerbate coral disease, but this relationship is likely complex as other factors also influence coral disease prevalence. To better understand this relationship, we meta-analytically examined 108 studies for changes in global coral disease over time alongside temperature, expressed using average summer sea surface temperature (SST) and cumulative heat stress as weekly sea surface temperature anomalies (WSSTAs). We found both rising average summer SST and WSSTA were associated with global increases in the mean and variability in coral disease prevalence. We showed global coral disease prevalence reached 9.92% compared to 3.16% in 1992, and the effect of ‘year’ became more stable (i.e., has lower variance), contrasting to the effects of the two temperature stressors. Regional patterns diverged over time and differed in response to average summer SST. Our model predicted that, under the same trajectory, 76.8% of corals would be diseased globally by 2100, even assuming moderate average summer SST and WSSTA. These results highlight the need for urgent action to mitigate coral disease. Mitigating the impact of rising ocean temperatures on coral disease alone is a complex challenge requiring global discussion and further study.


Introduction:
Host-pathogen-environment interactions over time, or "disease dynamics," are now an integral part of understanding ecosystem function in the context of climate change (Altizer et al., 2013;Burge et al., 2014;Vega Thurber et al., 2020). Diseases can drastically alter the composition and resilience of communities, which has been documented across ecosystems (Alvarez-Filip et al., 2022;Anderson et al., 2004;Burdon and Zhan, 2020;Burge et al., 2014;Estrada-Saldívar et al., 2020;Harvell et al., 2019). As increasing temperatures associated with climate change are a major driver of disease dynamics (Burge et al., 2014), climate change may expedite changes to biological communities by increasing the prevalence or severity of disease outbreaks (Aalto et al., 2020;Barris et al., 2018;Jones et al., 2021;Karvonen et al., 2010). For example, thermal extremes exert stress upon host immune systems and allow pathogens to emerge in new habitats, leading to higher disease rates (Burge and Hershberger, 2020;Byers, 2021;Harvell et al., 2007;Shields, 2019). Increasing temperatures may also change the timing or frequency of transmission periods, exposing vulnerable individuals to pathogens (Altizer et al., 2013). Furthermore, as temperature extremes are expected to increase in frequency, the time available for recovery after such events will be reduced, potentially contributing to the collapse of keystone species (Baker et al., 2008;Burge et al., 2014;Eakin et al., 2010).
Alternatively, pathogens also have thermal limits which may constrain their capacity to survive and reproduce, potentially restricting disease emergence or prevalence (Altizer et al., 2013;Byers, 2021). Therefore, understanding and predicting the impact of rising temperatures on disease dynamics is a challenging task and requires long-term global data on disease prevalence.
The sensitivity of different organisms to temperature-driven changes in disease dynamics can also depend on the type of ecosystem considered. Coral reefs are particularly sensitive to changes in sea surface temperature (SST), as shown by their susceptibility to bleaching (Glynn and D'Croz, 1990) and temperature-driven disease incidence (e.g., Walton et al., 2018;Bruno et al., 2007;Tracy et al., 2019;Randazzo-Eisemann et al., 2022;Howells et al., 2020). However, despite predictive models suggesting an upward trend in coral disease prevalence (e.g., Maynard et al., 2015;Walton et al., 2018), pathogen thermal limits and other factors may constrain coral disease increases (Altizer et al., 2013;Byers, 2021). Additionally, temperature fluctuations, such as heatwave events, experienced throughout life (i.e., a coral's "thermal life history") contribute to increasing the range of temperatures corals can tolerate, thus enhancing the ability to resist disease (Palumbi et al., 2014;Randall et al., 2014;Thomas et al., 2018;Ward et al., 2007). Therefore, thermal impacts on coral disease dynamics are highly complex.
Extrinsic and intrinsic factors other than temperature can also alter coral disease dynamics. For example, human activity (e.g., commercial overexploitation of marine ecosystems), water flow, and pollution have all been shown to influence coral disease (Lamb et al., 2014;Page et al., 2019;van de Water et al., 2015). Reef ecosystems with lower coral cover have displayed lower levels of disease prevalence, which is likely driven by limited pathogen transfer between such disparate individuals (Bruno et al., 2007;Caldwell et al., 2018;Dobbelaere et al., 2020;Zvuloni et al., 2015). Coral species with relatively slower population turnover rates exhibit higher levels of disease (Yakob and Mumby, 2011), hinting that species with a higher individual replacement rate may be less susceptible to disease. Concurrently, some coral populations appear more resistant to disease (Mydlarz et al., 2010), potentially due to differences in coral species assemblages, life histories, and environments (e.g., González-Barrios et al., 2021;Williams et al., 2021;Williamson et al., 2022). Therefore, there are likely to be regional differences in coral populations. Finally, the assessment of thermally-driven coral disease dynamics becomes even more complex over long time periods as the extrinsic human-induced factors are likely to concurrently increase over time (e.g., Lamb et al., 2014;Randazzo-Eisemann et al., 2022;Renzi et al., 2022;van de Water et al., 2015).
Progressing deterioration of coral communities and their key significance to marine life call for an urgent assessment of the expected magnitude and spatiotemporal trends of climate change-driven disease prevalence in corals. To do this, we conducted a global meta-analysis of disease prevalence over time in stony corals. We quantified the magnitude of the increase in coral disease prevalence over the last 40 years to determine the extent to which increases in average sea surface temperature (SST) and cumulative heat stress (as Weekly Sea Surface Temperature Anomaly -WSSTA) correlate with coral disease prevalence. Additionally, we determined how prevalence is likely to change independently of local rising temperatures -i.e., if factors other than temperature contribute to coral disease prevalence. Finally, we examined the global distribution of coral diseases to identify if regional characteristics account for some of the variations in coral disease prevalence of each ocean basin. This study marks the first compilation and analysis of coral disease surveys alongside SST records on a global scale to better understand how climate change continues to impact coral reef ecosystems. While coral reefs remain complex and diverse systems, deepening our understanding of global projections of coral disease dynamics will assist in developing effective conservation efforts with the intent of slowing or preventing the rise in coral diseases and maintaining the health and stability of these ecological and economical assets.

Reporting Guideline
We reported our study following PRISMA EcoEvo guidelines .
Our PRISMA diagram of literature search and screening ( Figure S1), as well as our PRISMA EcoEvo checklist, are available as Supplementary Materials. We also followed a systematic review method for literature search and selection, including piloting, benchmarking, and error checking (Foo et al., 2021).

Literature Search and Screening
We conducted a literature search in Scopus and Web of Science Core Collection in July 2020 using the piloted search string (Table S1). Our search string contained terms related to stony corals, disease, and climate change-induced temperature change. We tested the sensitivity of our search strategy against a set of 13 benchmark papers (Table S2). Besides searching the databases, we used two key reviews of the coral disease research (Montilla et al., 2019;Sokolow, 2009) to perform additional backward and forward reference searches. We used Rayyan (Ouzzani et al., 2016) to screen all 3,689 bibliographic records.
We screened the literature collection in two stages: abstract screening and fulltext screening. At both stages, we used pre-piloted decision trees ( Figure S2, S3) representing our selection criteria described below. Two reviewers (SB, PP) independently screened 150 randomly selected records to test the decision trees, yielding a 93% agreement rate between reviewers for abstract screening and 100% for the full-text screening stage. One reviewer (SB) then screened all remaining records.
This process resulted in 158 papers selected for data extraction (Table S3). After filtering and cleaning, our final dataset encompassed 108 papers for meta-analysis.
Papers excluded during the full-text screening stage can be found in Table S4.

Selection Criteria
We included studies based on six criteria. First, studies needed to have provided in situ empirical benthic survey data of coral disease, as surveys are a common method used for identifying coral and reef conditions. Second, surveys must have been on a natural reef (i.e., studied reefs were subject to natural environmental conditions). Third, surveys must have examined stony corals. Fourth, studies must have reported coral disease prevalence and relevant descriptive statistics (e.g., average or median of % prevalence) and sample size (number of assessed sample plots) for calculating effect sizes. Fifth, studies must have reported the year the survey was conducted. Sixth, studies must have been available in English (see more on this criterion in Discussion).
For consistency, we excluded studies that did not use transects, quadrats, or circle plots as the benthic survey collection method during data extraction (i.e., we excluded papers that used timed swims, etc., to survey reefs). We excluded these survey methods as quantifying the benthic area examined was unreliable or not possible to decipher from the paper, and study area was necessary for weighing each effect size (i.e., for conducting a weighted/formal meta-analysis).
Our second criterion excluded papers that examined corals from nurseries, laboratories, or other non-wild populations. However, we included studies conducted on corals from restoration efforts because these reefs do not differ from wild reefs in their exposure to environmental stressors (Afiq-Rosli et al., 2017;Monty et al., 2006;Rinkevich, 2014). In contrast, nurseries or other coral-growing facilities are generally protected from predators and other stressors, which could influence disease prevalence (Casey et al., 2014). Additionally, some reefs exist solely because of restoration efforts, such as the Acropora cervicornis populations in the Florida Keys (Miller et al., 2014), and excluding these would limit data coverage.
For the third criterion, we classified stony corals as reef-building corals with hard, calcareous skeletons. We focused specifically on stony corals for their ecological importance as habitat builders. Stony corals are also vital to coastal communities as they support both local economy and shore protection (Cesar and van Beukering, 2004). This criterion excluded papers which solely examined soft corals such as Alcyonacea since these corals differ from stony corals in which diseases they can acquire (Willis et al., 2004).
Included studies needed to have reported disease prevalence values as a percentage (or proportion), as per the fourth criterion. We define disease prevalence following Rogers (2010). Briefly, disease is typically associated with bacterial infection, discoloration of the tissue, and/or tissue loss. Discoloration and/or loss of tissue are the main identifiable signs used as evidence of disease presence in stony corals Raymundo et al., 2008;Rogers, 2010;Work and Aeby, 2011) and are used across the literature within benthic surveys to report coral disease prevalence. Thus, we used visual identification of disease signs as the primary reported method for identifying disease prevalence. Disease prevalence is a community-level metric, usually presented as the percentage/proportion of the coral community that displays symptoms (Rogers, 2010). Studies typically report overall disease prevalence across all stony coral taxa in the community, prevalence within the dominant taxa, and/or prevalence within the disease-impacted taxa on the reef. Prevalence differs from disease severity which measures disease progression rate within infected individuals (Rogers, 2010). While severity is an important metric for understanding how diseases will change with climate change, we did not examine it in this study and instead focus solely on prevalence as we found few studies consistently reporting severity during our pilot literature search.
We also excluded measurements reported as "bleaching" to disentangle colony-wide coral bleaching from coral disease signs (Rogers, 2010). When corals bleach, they eject their symbiotic zooxanthellae which does not meet our definition of disease.
Our fifth criterion required included papers to report the year(s) in which coral reef surveys were conducted. It was necessary for disease prevalence to be reported per year, even if the study examined a reef over many years. If multiple years were included in one prevalence metric, we determined that the effects of temperature would not be properly attributed to the prevalence observed. In cases where multiple years were conflated in one prevalence metric or the year was missing altogether, the paper was excluded.

Data Extraction and Coding
We used the proportion of disease prevalence as the effect size in this study. We incorporated the sampling error of this effect size through the sample area size, which we calculated using the number of sample plots and the area of each plot. We extracted the disease prevalence percentage, plot sample size ( In addition to our effect size statistic (% prevalence) and its weighting variable (sampling area), we collected data (i.e., moderators) in two different ways. Firstly, we directly gathered five variables from the included articles: 1) survey year (if conducted in the transition into a new year -e.g., the survey began in December and ended in January -we presented data as data collected in separate years -i.e., the prevalence of December and January separately -if able to do so; otherwise, we only utilized the year in which the survey started), 2) survey month(s), 3) the number of diseases identified during the survey encompassed in each effect size, 4) survey method (i.e., what type of transect/sample plot was used to map out a survey area), and 5) survey location (latitude and longitude; if not reported in the article, coordinates were estimated in Google Maps based on information provided in the papers such as figures/maps).
Secondly, we obtained three more variables using three types of external sources. These three variables are: i) survey region (based on Hoegh-Guldberg et al., 2017 andKleypas et al., 2008), ii) average summertime sea surface temperature (SST) in the summer prior to sampling in °C, and iii) the weekly sea surface temperature anomaly (WSSTA) measure for the sampling period in °C-weeks. The last two were calculated from SST databases available online, which are detailed below.
Survey locations were initially classified into ten regional locations (East Pacific, To analyze the effect of temperature on coral disease, we used two measures of temperature relevant to coral disease -SST and WSSTA (e.g., Bruno et al., 2007;Randall and Van Woesik, 2017). We used the average SST of the summer prior to sampling to investigate the influence of local average temperature change on disease prevalence. WSSTA measures the cumulative effect of anomalously high temperatures over a 52-week period and is thus used for identifying the influence of persisting anomalously high temperatures on disease prevalence (Bruno et al., 2007). While other metrics have been used to investigate heat stress on corals (e.g., degree heating week, hot snap, and cold snap -all usually reported with fine, i.e., day-to-day, resolution), WSSTA better corresponds to the yearly resolution of the extracted disease prevalence data. We also wanted to examine the global change in coral disease prevalence as a whole (i.e., not specific to any one disease), and metrics such as hot snap and cold snap require a finer geographical scale. They are also disease-specific as some diseases may be influenced by decreasing temperatures as opposed to increasing (Caldwell et al., 2016). Neither of these temperature metrics correlate significantly with Year, suggesting these represent the impact of local temperature stress ( Figure S4) and that global climate warming is still closely tied within the Year metric.
These two measures (average summer SST and WSSTA) were each obtained using different databases. We utilized a database from the National Oceanic and Atmospheric Administration Physical Sciences Laboratory online collection of gridded climate datasets to calculate average SST metrics during the summer months of the sampling year (Hirahara et al., 2014). We selected the COBE-SST2 dataset which included monthly SST means from January 1850 to December 2019 on a 1.0-degree latitude x 1.0-degree longitude global grid. Calculation of WSSTA was based on a database accessed through the Copernicus Climate Change Service (Lopez, 2019). We selected the "Sea surface temperature daily data from 1981 to present derived from satellite observations" dataset, compiled by The European Space Agency and Sea Surface Temperature Climate Change Initiative, which included daily SST measurements from January 1981 to the present day on a 0.05-degree latitude x 0.05degree longitude global grid, for its long history and high-resolution (Lopez, 2019). The higher resolution enabled us to more accurately calculate the cumulative heat stress that forms the WSSTA measure.
Average summer SST included average temperatures from June, July, and August for northern hemisphere surveys and December, January, and February for southern hemisphere surveys. Our decision to use temperature data from the summer prior to the sampling period stemmed from the evidence that a time lag occurs between anomalously high temperatures and visual signs of disease emergence (Caldwell et al., 2020;Heron et al., 2010;Maynard et al., 2011;Rudolf and Antonovics, 2005). If the year of sampling was used to calculate average SST, there would be concern that these temperatures would not truly correlate with the sampled disease if a time lag does occur.
WSSTA was calculated as the sum of positive deviations in weekly temperature averages from a threshold temperature over the 52-week period prior to disease surveys (Bruno et al., 2007). We set this threshold as 1°C greater than the maximum average monthly temperature in the 1981-1992 period (this period is commonly referred to in thermal stress calculations as "long-term climatology;" Figure S5). 1°C warmer than "long-term climatology" is the temperature at which corals start to experience thermal stress (Glynn and D'Croz, 1990). These thermal stress temperatures above the threshold during a 52-week window are summed together to produce the WSSTA value in units of "°C-weeks" representing the accumulation of weekly heat stress in the previous year-to-date period (Skirving et al., 2020).
We visually detail the data extraction and coding process in Figure S6 in Supplementary Materials. A list of extracted variables including additional variables, which we did not use in our analyses, and their descriptions are available in Table S5.

Statistical Analysis
We analyzed disease prevalence (proportion) weighted by the natural logarithm of sampling area (m 2 ), using a Bayesian zero-inflated generalized linear mixed-effects model (GLMM) with the beta-distribution family and without assuming homoscedasticity (i.e., explicitly modeling heteroscedasticity). This model was implemented in the R package brms (version 2.17.0; Bürkner, 2018Bürkner, , 2017. The beta distribution allowed us to model proportion data without underlying count data, and the zero-inflated distribution dealt with the presence of zeros, which cannot be formally modeled under a beta-family GLMM. This GLMM used a logit link function for the main (beta distribution, denoted as "mu") and zero-inflated (Bernoulli distribution, denoted as "zi") parts while the log link function was used to model the precision (log-normal distribution, denoted as "phi"), which represents the degree of heteroscedasticity.
Our GLMM included the following five fixed effects: 1) weekly sea surface temperature anomaly (WSSTA), 2) average summer sea surface temperature (SST), 3) year at the start of the survey, 4) ocean (i.e., Pacific, Atlantic, and Indian), and 5) the number of diseases identified. All the continuous variables were scaled for interpretability (Schielzeth, 2010). The model also had the following four random effects: i) site identity (unique locations; 199 levels), ii) paper identity (108 levels), iii) season at the start of the survey, adjusted for hemisphere (4 levels), and iv) method of data collection (i.e., belt transects, quadrats, belt and quadrat, circle plots, and line transects; 5 levels). The first two random effects deal with non-independence among effect sizes.
This GLMM also incorporated the correction for zero-inflation (zi) and precision (phi) described above which were modelled using the three key predictors: WSSTA, SST, and Year. This GLMM constituted our base model.  Table S6). Regional differences were compared using pairwise estimates with each ocean basin. All the models were run with the default prior, iteration = 30,000, and warm-up = 28,000 with two chains, which gave us an effective posterior sample size over 4,000 for all the parameters. We considered regression coefficients statistically significant if 95% credible intervals did not overlap with zero.
Since magnitude cannot be identified when relationships are non-linear, we examined if the rates of change in disease prevalence were significantly different between the maxima and minima (i.e., the extreme values) for each predictor that had a non-linear relationship with coral disease prevalence. We conducted this comparison using the emtrends function of the emmeans package (version 1.7.3; Lenth, 2022). Predictions using our model are further complicated by the fact that the year effect will inevitably contain the effects of the global trend in rising temperatures. Therefore, it is impossible to separate the time trend from the global warming trend completely. Our predictions into the future should always be seen as projecting the prevalence along the current (a linear "business-as-usual") global warming trajectory. In this context, the average SST predictor should be seen as the finer and more local effect of temperatures captured at a smaller spatial scale.

Characteristics of Disease Surveys and Literature
Our dataset comprised 108 papers which yielded 918 effect sizes. A visual summary of the literature screening is presented as a PRISMA diagram ( Figure S1). Table S3. Excluded papers from full-text screening are listed in Table S4. Our data included coral surveys conducted between 1992 and 2018.

Included papers are listed in
About half of the data collected was surveyed in the Atlantic Ocean (50.5% of effect sizes), with fewer surveys from the Pacific Ocean (35.1%) and Indian Ocean (14.4%) ( Figure 1A). Most surveys began during summer months: 48.1% of effect sizes in the northern hemisphere, and 38.7% in the southern hemisphere ( Figure 1B). The most surveyed disease was White Syndrome (30.4%), followed closely by Black Band Disease (30.0%) and Yellow Band Disease (22.3%) (Table S7). Many studies reported disease prevalence per disease identified (78.5%), but some papers did not split disease prevalence into measures of a single disease and present an aggregated disease prevalence for all diseases identified (21.5%, Figure 1C). For information regarding the coral species examined and effect size distribution for each fixed factor, see supplementary information ( Figure S7, Supplementary Materials).

The Effect of Average Summer Sea Surface Temperature (SST)
In our base model without interactions with Ocean (i.e., region), a rise in average summer SST predicted a non-linear increase in coral disease prevalence (Figure 2A-B).

The Effect of Weekly Sea Surface Temperature Anomaly (WSSTA)
Our base model showed that increasing WSSTA predicted a non-linear increase in disease prevalence ( Figure Figure 2F, Figure S8B). In the base model, this increase consisted of three parts: 1) a significant increase in nonzero disease prevalence observations (b = 0.2, 95% CI = 0.16 to 0.23; Figure 3D), 2) a significant decrease in zero disease prevalence observations (i.e., fewer instances of coral observed without any signs of disease; b = -0.17, 95% CI = -0.24 to -0.11; Figure   3E), and 3) a significant decrease in precision for disease prevalence (i.e., disease prevalence became less predictable as WSSTAs increased; b = -0.24, 95% CI = -0.26 to -0.22; Figure 3F).

Regional Differences
To test for regional differences, we conducted model selection among models that included interactions between the three oceans (Ocean: Pacific, Atlantic, Indian) and our three key predictors (average summer sea surface temperature -SST, weekly sea surface temperature anomaly -WSSTA, and Year). The best fitting interaction model contained interactions between Ocean and average summer SST, and Ocean and Year (Table S6). In this model, the difference between the slopes of these three oceans for average summer SST were significant for Atlantic-Pacific and Indian-Pacific  Figure S9A), but differences were non-significant for the Atlantic-Indian interaction (b[Atlantic-Indian] = -0.05; 95% CI = -0.28 to 0.16; Figure S9A). The difference between slopes among these three oceans for Year were also significantly different for the Indian-Pacific and Atlantic-Pacific interactions (b[Indian-Pacific] = -0.62; 95% CI = -1.11 to -0.14; b[Atlantic-Pacific] = -0.24; 95% CI = -0.49 to -0.02; Figure S9B) and were nonsignificant for the Atlantic-Indian interaction (b[Atlantic-Indian] = 0.38; 95% CI = -0.06 to 0.82; Figure S9B). Yet, the three regions showed similar patterns in disease prevalence in relation to the three predictors (average summer SST, WSSTA, and year), except for the disease prevalence observed through time in the Indian Ocean ( Figure 4).

Discussion
We conducted the first-ever global meta-analysis to quantify long-term trends in the magnitude of coral disease prevalence and determine the extent to which sea surface temperature (SST) or other factors influence coral disease prevalence. Average summer SST, Weekly Sea Surface Temperature Anomaly (WSSTA), and Year all significantly positively correlated with coral disease prevalence. We also newly discovered that these variables significantly influenced the predictability of disease prevalence (Figure 3). In addition, the Pacific Ocean significantly differed from the Atlantic and Indian Oceans regarding the effects of average summer SST and Year, although the directional patterns were consistent across different regions, with one exception (Indian Ocean disease prevalence showed a non-significant declining trend with Year; Figure 4). Below, we discuss the three key moderators (average summer SST, WSSTA, and Year) in turn and further elucidate the regional differences.

Seemingly contradictory effects of average summer SST
We found that coral disease prevalence increased with rising average summer SST (Figure 2, Figure 5A). Such a pattern was to be expected, as numerous studies have supported this trend (e.g., Bruno et al., 2007;Hazraty-Kari et al., 2021;Howells et al., 2020;Walton et al., 2018). In fact, when isolating the effect of these rising local summer temperatures, we expect coral disease to more than double by 2100 (19.6%, 95% CI = 5.5% -41.1%; compared to 2018: 9.92%, CI = 2.08% -24.5%). However, the prevalence of apparently healthy corals (no observed disease; zero-inflated component of the model; Figure 3) also increased with increasing average summer SST. This rise in healthy corals with increased SST directly contradicts previous literature (e.g., Walton et al., 2018).
There were hundreds, if not thousands, of coral species included in our global meta-analysis. Each species may react to changes in average summer SST differently, which might account for an increase in zero-disease (i.e., 'healthy') observations. While such observations do not negate the prediction of the overall increase in coral disease with rising summer SSTs, it is important to consider the different responses and resistance capacities to thermal stress among coral species (Drury, 2020;Guest et al., 2012;McClanahan et al., 2020). For example, reefs in the Persian-Arabian Gulf experience high temperatures and larger temperature variability (Camp et al., 2018). As the reefs in the Persian-Arabian Gulf are younger, they are hypothesized to be more able to withstand the current extremes of their environment (Camp et al., 2018).
Through these varied responses to thermal stress, these coral species may dominate the observations of zero disease prevalence in future surveys.
We further found greater variability in disease prevalence with increasing average summer SST (Figure 3, Figure 5B). Such an increase in variability has never been formerly reported before, yet this can be explained by varying responses by different coral species, as with the rise in disease-free corals observed. In addition, coral reefs are complex habitats, and there is substantial variation in bleaching and heat stress responses that occur within and between reefs during bleaching events Fordyce et al., 2021;Page et al., 2019). A rise in disease prevalence variation with increasing average summer SST indicates that it will be more difficult to predict disease prevalence as average summer SST rises. As a result, we may underestimate the severity of rising local SST on coral disease and fail to act within the available timeframe to conserve coral reefs.

The harmful effects of heat stress (WSSTA)
Our model suggested that as WSSTA (i.e., annually accumulated heat stress) increases, disease prevalence increases and the fraction of corals observed without disease symptoms decreases ( Figure 3E). Therefore, increasing WSSTA was associated with higher rates of disease overall, which indicates heat stress is likely linked with coral disease. This is consistent with a study conducted by Bruno et al., (2007) where they found that annually accumulated heat stress was significantly correlated with an increase in white syndrome. However, Bruno et. al. (2007) noted that high coral cover influenced disease prevalence associated with WSSTA. While we were unable to determine coral cover across all effect sizes, since we also found WSSTA correlates with high coral disease, our data was most likely collected using densely populated samples.
The identified increase in coral disease prevalence with WSSTA in the current study is consistent with previous studies of coral disease and heat stress (Aeby et al., 2021;Eakin et al., 2010). As coral disease appears highly correlated to accumulated heat stress, without mitigation, it is likely that high disease prevalence will yield greater coral mortality. We also found disease prevalence becomes more variable (i.e., precision decreased; Figure 3F, Figure 5). The increasing variability of disease occurrence with rising WSSTA, as with average summer SST, highlights once again the difficulty in predicting disease prevalence.

Coral disease through time
We predicted that coral disease prevalence will increase in future years ( Figure   2) with consistently most corals bearing visible symptoms of disease ( Figure 3I). By 2100, our model predicts 76.8% of corals in reefs will be infected globally (95% CI = 53.2% -92.9%), provided that WSSTA and average summer SST do not exceed their averages (2.08 o C-weeks and 28.6 o C, respectively). This prediction represents the IPCC "business as usual" RCP 8.5 climate projection (IPCC, 2022;Riahi et al., 2011). As local temperature stressors (i.e., WSSTA and average summer SST) increase, we expect even greater disease prevalence in reefs. For example, if average summer SST reaches 32.0 o C, we predict 80.5% (95% CI = 64.4% -92.4%) disease in 2100 (Logan et al., 2014).
Our model predicted a consistent increase in coral disease occurrence even after accounting for the detrimental effects of rising temperatures in the form of WSSTA and average summer SST (Figure 3, Figure 5). In particular, predictions of later years at the mean temperature conditions yield a greater change in coral disease prevalence than predictions in the same year at higher temperature conditions. This latter prediction clearly reflects that additional factors, other than the two thermal conditions we examined, are at play in driving coral disease worldwide. Although we could not account for all additional factors explicitly in our models due to data gaps and the heterogenous nature of reporting other potentially important variables, the year effect provides a rough proxy of their combined effect as these factors are expected to increase with time.
Some of the most commonly examined factors, apart from thermal factors, include ocean acidification (Prada et al., 2017), pollution (Redding et al., 2013), and anthropogenic damage due to, e.g., intrusive tourism practices (Lamb et al., 2014). Our synthesis clearly emphasizes a need for more disclosure of accompanying stressor variables and calls for the standardization of disease prevalence and stressor reporting.
Future studies should strive to scrutinize these stressors such that specific influences can be identified and mitigated.

Differences and similarities among the oceans
Overall directional patterns of coral diseases across the three oceans were largely similar concerning the three variables: average summer SST, WSSTA, and Year ( Figure 4), except for the correlation between coral disease and year in the Indian Ocean. This consistency is most likely because temperature rises have occurred globally with all oceans intaking heat-increasing radiation (Cheng et al., 2019).
However, year and average summer SST, rather than thermal stress (WSSTA), are the variables that seem to be driving regional differences (Figure 4). While all oceans can expect an increase in disease prevalence concerning rising summer SST, the Pacific Ocean is predicted to experience the slowest and steadiest increase in disease prevalence ( Figure 4C). In contrast, the Pacific Ocean is predicted to experience a more severe increase in disease prevalence than the Indian or Atlantic Oceans, independently from the two temperature-related conditions ( Figure 4H,I).
Several factors may underlie these regional differences in the effect of our three moderators on coral disease prevalence. Within each region, coral species are found in varying abundances and diversity. Coral species differ in heat tolerance ranges (e.g., -Guldberg O. et al., 2007) and in symbionts that could aid in thermal tolerance (Santoro et al., 2021). As stress induced by factors other than temperature (e.g., acidification, pollutants) can lower immune response , differences in stress resistance may contribute to observed differences in disease dynamics between oceans. Additionally, the stressors themselves are heterogeneous spatially, as many of these stressors (e.g., pollutants, overfishing, tourism) are anthropogenic (Vega Thurber et al., 2020). The predicted rise in coral disease in the Pacific Ocean over time (independently from temperature) suggests that factors unrelated to temperature, such as tourism and acidification, most likely heavily influence coral disease in this region.

Limitations and recommendations
Our analysis represents the first systematic synthesis of global coral disease data. Firstly, the available global datasets for coral disease are sparse, both temporally and geographically (i.e., the data is highly concentrated to particular reefs, especially in the Caribbean area of the Atlantic Ocean), potentially creating biased predictions for certain years and locations. Second, we focused solely on publications in English due to logistic limitations. While our literature sample size is large, assessing more regions (given coral reefs are in developing nations) and including as many languages as possible in future reviews would aid in accurately describing the state of knowledge and incorporate a more global representation of data. Other languages make up 35% of literature in similar fields (Amano et al., 2016) and recent reports suggest non-English literature does augment environmental data in a non-negligible way Pottier et al., 2022). Moreover, we also excluded grey literature, such as government reports. However, the inclusion of grey literature can create potential bias in the data from the lack of rigorous assessment of research quality (Bostrom-Einarsson et al., 2020). As such, this seeming limitation may have improved the quality of data synthesized in our study, which can be tested in future analyses.
Our synthesis has revealed four improvements that future primary studies could adopt to increase the relevance, visibility, and comparability of their data. Firstly, comparative studies would benefit from the complete adoption of standardized methodologies. Belt transects are the most common collection method within the literature of the past four decades (Teague et al., 2022). Second, prevalence reported per disease will also aid in data comparison, as most diseases are visually distinct and known to be linked to different drivers (e.g., water pollution and black band disease, heat stress and white syndrome; Teague et al., 2022;Weil and Rogers, 2011;Willis et al., 2004). Third, reporting the abundance of corals per site alongside disease percentages would allow researchers to better conduct meta-analyses, as it is necessary to distinguish between data collected from a reef with many individuals (i.e., high live cover) and a reef with few individuals (i.e., poor cover; Jameson et al., 2001).
Fourth, there is a need for more research across a greater range of locations to ensure a more geographically robust understanding of coral disease. The Caribbean is highly studied as it was impacted by disease during the past four decades and has a high research presence in the region (Morais et al., 2022), whereas little data exists for many Indian Ocean reefs. We expect these improvements to enable better comparisons for systematic reviews to understand global trends and drivers of coral disease.
In conclusion, as coral disease is expected to rise in future years (76.8% of corals diseased by 2100), it is imperative to identify drivers of coral disease. Our study is the first step towards paving the way for policymakers to develop effective mitigation strategies specific to these risks in their respective regions. Our meta-analysis highlights the devastating impacts of rising temperatures on coral reefs and the dire need for swift action to mitigate climate change.

Author Contributions:
Conceptualisation

Conflict of Interest Statement:
Authors have no competing interests to declare.

Acknowledgements:
We thank Scott Heron for his advice on sea surface temperature databases. SB and PP appreciate the financial support provided by the UNSW Scientia Doctoral Scholarship. SMD acknowledges the support of the ARC Discovery Early Career Award (DE180100202). SN, ML, and ELM were supported by the ARC Discovery grant (DP200100367).

Data and Code Availability:
Data compiled during the main study is available in a GitHub repository (https://github.com/sburke-unsw/CoralDiseaseMetaAnalysis). Data are stored in .csv format with a meta-data table in the Supplementary Materials. Code generated in this study is also available in the GitHub repository and provided as .R or .Rmd files.  Ocean colored in teal blue, and Pacific Ocean colored in yellow-green. Black trend line