The global, regional, and national burden of gastro-oesophageal reflux disease in 195 countries and territories, 1990-2017: a systematic analysis for the Global Burden of Disease Study 2017.

BACKGROUND
Gastro-oesophageal reflux disease is a common chronic ailment that causes uncomfortable symptoms and increases the risk of oesophageal adenocarcinoma. We aimed to report the burden of gastro-oesophageal reflux disease in 195 countries and territories between 1990 and 2017, using data from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017.


METHODS
We did a systematic review to identify measurements of the prevalence of gastro-oesophageal reflux disease in geographically defined populations worldwide between 1990 and 2017. These estimates were analysed with DisMod-MR, a Bayesian mixed-effects meta-regression tool that incorporates predictive covariates and adjustments for differences in study design in a geographical cascade of models. Fitted values for broader geographical units inform prior distributions for finer geographical units. Prevalence was estimated for 195 countries and territories. Reports of the frequency and severity of symptoms among individuals with gastro-oesophageal reflux disease were used to estimate the prevalence of cases with no, mild to moderate, or severe to very severe symptoms at a given time; these estimates were multiplied by disability weights to estimate years lived with disability (YLD).


FINDINGS
Data to estimate gastro-oesophageal reflux disease burden were scant, totalling 144 location-years (unique measurements from a year and location, regardless of whether a study reported them alongside measurements for other locations or years) of prevalence data. These came from six (86%) of seven GBD super-regions, 11 (52%) of 21 GBD regions, and 39 (20%) of 195 countries and territories. Mean estimates of age-standardised prevalence for all locations in 2017 ranged from 4408 cases per 100 000 population to 14 035 cases per 100 000 population. Age-standardised prevalence was highest (>11 000 cases per 100 000 population) in the USA, Italy, Greece, New Zealand, and several countries in Latin America and the Caribbean, north Africa and the Middle East, and eastern Europe; it was lowest (<7000 cases per 100 000 population) in the high-income Asia Pacific, east Asia, Iceland, France, Denmark, and Switzerland. Global prevalence peaked at ages 75-79 years, at 18 820 (95% uncertainty interval [95% UI] 13 770-24 000) cases per 100 000 population. Global age-standardised prevalence was stable between 1990 and 2017 (8791 [95% UI 7772-9834] cases per 100 000 population in 1990 and 8819 [7781-9863] cases per 100 000 population in 2017, percentage change 0·3% [-0·3 to 0·9]), but all-age prevalence increased by 18·1% (15·6-20·4) between 1990 and 2017, from 7859 (6905-8851) cases per 100  000 population in 1990 to 9283 (8189-10 400) cases per 100  000 population in 2017. YLDs increased by 67·1% (95% UI 63·5-70·3) between 1990 and 2017, from 3·60 million (1·93-6·12) in 1990 to 6·01 million (3·22-10·19) in 2017.


INTERPRETATION
Gastro-oesophageal reflux disease is common worldwide, although less so in much of eastern Asia. The stability of our global age-standardised prevalence estimates over time suggests that the epidemiology of the disease has not changed, but the estimates of all-age prevalence and YLDs, which increased between 1990 and 2017, suggest that the burden of gastro-oesophageal reflux disease is nonetheless increasing as a result of ageing and population growth.


FUNDING
Bill & Melinda Gates Foundation.


Introduction
Gastro-oesophageal reflux disease is a common and usually chronic ailment of the upper digestive tract. Some reflux of stomach contents into the oesophagus, with or without symptoms, is physiological. Gastrooesophageal reflux disease, however, is defined as a condition that develops when the reflux of stomach contents causes troublesome symptoms, complications, or both. 1 Why some individuals have more frequent or severe symptoms or complications of reflux than others is poorly understood, but obesity, hiatal hernias, alcohol, smoking, and various foods and medications have been reported as risk factors. [2][3][4] A positive association with age has been observed in many 4 -but not all 5 -studies.
Gastro-oesophageal reflux disease syndromes include typical reflux (defined by heartburn, regurgitation, or both, and sometimes accompanied by belching, water brash, or nausea), angina-mimicking chest pain, and extra-oesophageal symptoms such as chronic cough and chronic laryngitis. 1,6 Complications of gastro-oesophageal reflux disease include oesophageal inflammation, stricture, 7 ulceration, perforation, metaplasia (ie, Barrett's oesophagus), and oesophageal adenocarcinoma. [8][9][10][11][12] Associations of varying strength have been detected between reflux beyond the oesophagus and outcomes such as dental erosion, 13 difficulty controlling concurrent asthma, 6,14 and increased risk of laryngopharyngeal carcinoma. 15 Lifestyle changes to reduce reflux of stomach contents, such as weight loss and eating smaller meals, are commonly recommended (eg, by treating physicians and in practice guidelines written by professional organisations and committees) and moderately supported by evidence. 16,17 Often, however, effective control of symptoms requires the use of acid-suppressing medications, such as protonpump inhibitors. Long-term use of proton-pump inhibitors has been associated with adverse outcomes such as loss of bone-mineral density and increased occurrence of enteric and pulmonary infections. [18][19][20][21][22][23][24][25][26][27] Surgical or endoscopic proced ures to reduce reflux are done in selected medication-dependent or refractory cases. Health-care systems and individuals incur economic costs for physician visits, medications, and procedures. [28][29][30][31][32][33][34][35] Objective measures such as oesophageal pH monitoring or endoscopy can be used to diagnose gastro-oesophageal reflux disease or its effect on oesophageal mucosa, but these procedures are invasive and can miss cases with fluctuating course. Multiple expert groups have endorsed the use of clinical history and response to therapy in making a clinical diagnosis. 1,36 Multiple symptom-based questionnaires have been developed for use in populationbased research, 37 and prevalence studies have mainly been carried out with this approach.
Several systematic reviews have been published in the past two decades describing the incidence and prevalence of gastro-oesophageal reflux disease. 38,39 The methodology of systematic reviews, however, limits comparisons across geography and time to those geographies and times for which reported studies exist, and does not quantitatively account for differences in study design. Eusebi and colleagues did a meta-analysis of gastro-oesophageal reflux disease, 4 which produced global and regional pooled estimates of disease prevalence and explored features of study designs that might explain inter-study heterogeneity, but did not use information about these design features to adjust the

Research in context
Evidence before this study The Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) has not estimated the burden of health loss due to gastro-oesophageal reflux disease. Two previous systematic reviews and one previous meta-analysis evaluated the prevalence of gastro-oesophageal reflux disease and its geographical variation. These studies suggested that the prevalence of this disease around the world ranged from 2·5% to 33·1%, and that prevalence was lower in east Asia and southeast Asia. One systematic review suggested that prevalence increased after 1995. The designs of these studies did not quantitatively account for the effect that differences in study design might have on study results, and only provided estimates of prevalence for the small number of countries where original studies have been done or for broadly defined regions, and did not estimate the burden of gastrooesophageal reflux disease in terms of years lived with disability (YLDs) or other composite measures of health loss.

Added value of this study
GBD 2017 provides the first comprehensive estimates of global, regional, and country-specific prevalence and non-fatal health loss due to gastro-oesophageal reflux disease for 195 countries and territories, from 1990 to 2017, using patterns observed in data from different locations, ages, and times to produce the best possible estimates both where data are available and where they are not. GBD 2017 incorporated more data sources on the prevalence of gastro-oesophageal reflux disease than previous systematic reviews and meta-analyses, and used a modelling approach that adjusted for the effects of nonstandard study designs on prevalence data. Even after these adjustments, GBD 2017 generally confirmed the findings reported in previous studies with regard to the range of gastrooesophageal reflux disease prevalence seen worldwide and the finding that prevalence is lower in countries in east Asia and in the high-income Asia Pacific, but it did not find a global increase in the prevalence of gastro-oesophageal reflux disease after accounting for population ageing.

Implications of all the available evidence
Gastro-oesophageal reflux disease is common and increasing due to population ageing. Health-care systems should be prepared to address the needs of increasing numbers of patients with gastro-oesophageal reflux disease. In some locations, there might be an increase in the prevalence of gastro-oesophageal reflux disease beyond the increase due to age, but more research is required to determine whether this is true and, if so, what factors are driving this increase and what interventions might decrease the burden of gastro-oesophageal reflux disease. contribution of non-standard studies to pooled estimates. Furthermore, the chronicity of gastrooesophageal reflux disease and the fact that it can cause persistent or episodic symptoms of varying severity make it important to move beyond estimations of incidence and prevalence, and to quantify the severity and duration of health loss it causes. The Global Burden of Disease research framework uses meta-regression methods to synthesise data from published studies to make estimates for 195 countries and territories worldwide from 1990 to the present, and expresses the relative health loss due to more than 350 diseases and injuries in common terms that facilitate comparisons. Here, we report results from the Global Burden of Diseases, Injuries, and Risk Factors Study (GBD) 2017, the first iteration of GBD to estimate non-fatal health loss due to gastro-oesophageal reflux disease.

Overview
The overall objectives, methods, and organisation of GBD 2017 have been previously reported. [40][41][42] Methods relevant to estimating the burden of gastro-oesophageal reflux disease are summarised here and described further in the appendix (pp 1-8).
For our analysis, individuals with heartburn, regurgitation, or both, at least once weekly over a 12-month recall period, were defined as having gastrooesophageal reflux disease. This definition was chosen over the consensus-group-recommended definition of mild symptoms occurring at least twice a week or moderate to severe symptoms occurring at least weekly 1 because of greater data availability, and is consistent with a previously published meta-analysis. 4 Individuals who had oesophageal complications (eg, ulceration or meta plasia) without symptoms, whose sole symptom of gastro-oesophageal reflux was chest pain without typical reflux symptoms, or who had reflux primarily as a trigger or exacerbating factor in respiratory or head and neck diseases (eg, chronic cough or dental erosion) were not included. This strategy avoids double-counting disability already attributed to other underlying diseases modelled in GBD.

Prevalence estimation
Data inputs for estimating the prevalence of gastrooesophageal reflux disease included epidemiological studies of gastrointestinal illness published in peerreviewed journals and identified in a systematic review via PubMed, and data from the US National Health Interview Surveys. Search terms and other details of the systematic review are provided in the appendix (pp 1-3). A complete set of unadjusted input data included in the model can be downloaded from the GBD 2017 Data Resources website. Extracted data from studies with acceptable but non-preferred designs were marked with study-level covariates to allow for estimation of fixed effects due to study characteristics in our global metaregression analysis (described later).
Gastro-oesophageal reflux disease data were analysed with a Bayesian mixed-effects meta-regression framework, DisMod-MR 2.1, developed for GBD non-fatal estimation processes, which has been previously described in detail [42][43][44] and is summarised here. Estimates are made by fitting a series of models, each of which serves to generate a Bayesian prior distribution for a subsequent model. At each step, DisMod assumes a compartmental disease model with three states-susceptible, diseased, and dead-with transition between states determined by incidence, remission, excess mortality due to disease, and other-cause mortality. These disease parameters are modelled with an offset log-normal data likelihood function, and a system of age-integrated differential equations are solved to ensure internal consistency among disease parameters.
The first model in the DisMod series is a global mixedeffect model, which uses all data from both sexes, all locations, and all years, and estimates coefficients for fixed effects for sex, study design characteristics, and predictive covariates, and random effects for each of the seven GBD super-regions. The next step is to fit separate mixed-effects models for each year, sex, and superregion, each of which re-estimates the fixed effect coefficients and estimates random effects for each GBD region within that super-region; the Bayesian prior distribution for each super-region-level model is based on the distribution estimated by the initial global model with the fixed effects and the random effect for that super-region. This method is repeated to fit separate mixed-effects models specific to sex, year, and region, using the preceding super-region model and the random effect for the region to determine the Bayesian prior, and estimating random effects for countries. This approach is again repeated to fit separate models specific to sex, year, and country, using the preceding regional model and the random effect for the country to determine the Bayesian prior. For 15 countries, an additional round of models is fit for subnational units (such as states or provinces), each deriving its Bayesian prior from its country model and a pseudo-random effect based on the average ratio of observed subnational data to countrymodel predictions. This algorithm for developing prior distributions for subnational models is sensitive to data in age groups that have low estimated values in the country-level fit, which can cause the model to ignore the preponderance of the data; in these cases, data for the affected age groups in the subnational locations are excluded.
As mentioned, the DisMod framework estimates fixed effects for study design characteristics; these study-level fixed effects reflect the association observed in input data between study design characteristics and measured disease parameters, and they serve to adjust for measurement bias due to non-reference study designs. Fixed effects are also estimated for predictive covariates; these reflect the association observed between that covariate and disease input data and serve to help estimate disease parameters in locations with scarce or absent input data. To be considered as a predictive covariate, a factor must have a demonstrated association with disease in non-GBD studies, and valid estimates of the distribution of that factor must exist for all GBD locations and estimation years available to use as DisMod inputs. 42 The association between a predictive covariate and disease parameters need not be causal to serve this purpose. Candidate predictive covariates found to have null or highly uncertain coefficients in preliminary models do not improve estimates, so they are left out of the final model for parsimony.
Ultimately, final estimates for national or subnational locations reflect local data, adjusted for study design characteristics, if local data are present, and reflect prior distributions from broader geographical units and the influence of predictive covariates if no local data are available. Estimates from the finest level of geography are later aggregated to make final estimates for the broader geographical units. Uncertainty intervals are taken as the 2·5th and 97·5th percentiles of the posterior distribution.
Parameters used in DisMod for gastro-oesophageal reflux disease were as follows: excess mortality was assumed a priori to be 0, and remission prior was set to 0·2-0·5 cases per person-year. Incidence was forced to 0 from birth to age 5 years, and after this age prior was set to 0·0-0·2 cases per person-year. We included studylevel covariates for alternative recall periods, for alternative minimum symptom frequencies, for the use of a score-based case definition that synthesised the severity, number, and frequency of symptoms, for the use of a case definition based on a single cardinal reflux symptom (regurgitation only), for studies in which the representativeness of the sample was considered questionable, and for data extracted from a report from a national survey, rather than a peer-reviewed publication. We considered location-level covariates for mean bodymass index (BMI), smoking prevalence, mean alcohol consumption, 45 and the Healthcare Access and Quality Index, 46 but these covariates were non-predictive in preliminary models, so they were not retained in the final model.

Estimation of years lived with disability
Years lived with disability (YLDs) synthesise the frequency and non-fatal health consequences of a disease. YLD estimation in GBD 42 begins by estimating the point prevalence, specific to year, age, sex, and location, of specific health states that can result from the disease, generally at different levels of severity. Each of    these disease states corresponds to one of a set of health states for which disability weights have been derived from population-based surveys. [47][48][49] Health states describe the consequences of disease or injury in terms relevant to an individual's life, such as loss of function and pain or other symptoms. The disability weights for these health states range from 0 to 1, with 0 representing perfect health and 1 representing death. Prevalent cases in each health state are multiplied by the disability weight of that health state to calculate YLDs. In a microsimulation process, all health states for all diseases are assigned to simulants according to their point-prevalence specific to year, age, sex, and location, assuming independent probability. For simulants assigned health states for multiple diseases, YLDs are adjusted with a multiplicative function of the disability weights. YLDs due to all health states of each disease are then summed.
The prevalence of health states for gastro-oesophageal reflux disease was determined from severity and frequency distributions reported in the prevalence studies used in our prevalence model. Severity and frequency categories were combined, as described below, to generate four categories, and these categories were assigned the health states and disability weights shown in the appendix (p 5).
Throughout the literature, the severity of gastrooesophageal reflux disease is often divided into two to five categories according to diverse definitions. We reviewed the studies in our input data and, if provided, extracted counts of cases of each severity as reported. These cases were then mapped to one of two GBD 2017 gastro-oesophageal reflux disease severities: mild to moderate (disability weight 0·011; 95% uncertainty interval [95% UI] 0·005-0·021) and severe to very severe (disability weight 0·027; 0·015-0·046). [47][48][49] The proportion of cases in each of the GBD 2017 gastro-oesophageal reflux disease severities was calculated for the pooled total cases, along with standard errors based on a simple proportion model.
Many studies also report the frequency of gastrooesophageal reflux disease symptoms as the proportions of cases in each of a set of mutually exclusive and collectively exhaustive frequency categories. Examples include 1-6 days per week and daily; 1 day per week, 2-6 days per week, and daily; 1-3 days per week, 4-6 days The super-regions North Africa and the Middle East and South Asia each contain only one region, which bears the same name, so these rows are not repeated. 95% UI=95% uncertainty interval.

Table:
Prevalence of gastro-oesophageal reflux disease in 1990 and 2017 for both sexes and all locations, with percentage change per week, and daily; and so on. For each study, 1000 proportion draws were generated for each frequency category with a beta distribution. These proportion draws were multiplied by the assumed mean days per week symptomatic for the category (the midpoint of the range) to produce draws of the number of days per week symptomatic that were contributed by cases in that category, and these draws for proportion-weighted means were summed across categories to estimate days per week symptomatic for all cases in the study. Means and SDs of these draws were combined in a meta-analysis, and the final mean and SD were divided by seven to estimate the proportion of cases that were symptomatic on a given day, with uncertainty.
Data about severity and frequency were too sparse to adjust meta-analyses for person, place, or time, so the same pooled proportions were applied to all combinations of year, age, sex, and location.
Because a single distribution of severity and frequency was applied to calculate YLDs for all years, ages, sexes, and locations, all variation in YLDs is driven by variation in prevalence. Because fatalities related to gastrooesophageal reflux disease are attributed to other underlying causes of death (eg, oesophageal carcinoma), no years of life lost (YLLs) are directly estimated for gastrooesophageal reflux disease and disability-adjusted lifeyears (DALYs) are equal to YLDs.
Final estimates of prevalence and YLDs were specific to year, age, sex, and location. These estimates were weighted and aggregated by the age-sex distribution of the population in the location and year to which the estimates applied to produce all-age estimates. The same year-agesex-location-specific estimates were adjusted to the GBD reference population by direct methods as previously described to produce age-standardised estimates. [50][51][52] The percentage change in estimates between 1990 and 2017 was estimated by calculating the percentage change between pairs of 1000 draws from the bootstrap distributions of estimates for each year, then finding the mean and 25th and 975th ordered values of the resulting combined distribution.
At the recommendation of GBD network collaborators, as a post-hoc analysis, final age-standardised YLD rate estimates were plotted against GBD estimates of Sociodemographic Index 42 and their relationships modelled with reduced cubic splines.
We documented each step of the GBD 2017 estimation processes, as well as data sources, in accordance with the Guidelines for Accurate and Transparent Health Estimates Reporting (GATHER) statement.

Role of the funding source
The funder of the study had no role in study design, data collection, data analysis, data interpretation, or writing of Age-standardised prevalence (per 100 000) 4400 to <5000 5000 to <6000 6000 to <7000 7000 to <8000 8000 to <9000 9000 to <10 000 10 000 to <11 000 11 000 to <12 000 12 000 to <13 000 13 000 to <14 035 the report. The corresponding authors had full access to all the data in the study and had final responsibility for the decision to submit for publication.

Results
In our systematic review, we found 112 studies that met the inclusion criteria. Four studies used diagnostic codes to identify cases in administrative data, two studies used self-reported diagnosis, and the remainder were surveys that used symptom-based questionnaires: 27 studies used the GBD case-definition for gastrooesophageal reflux disease, and 79 studies used one of more than 50 alternatives that differed in recall period, minimum symptom frequency, defining symptoms, or manner of scoring. Combined with data from a household survey, this strategy provided 144 locationyears of prevalence data and 406 prevalence datapoints; six datapoints in young age groups in subnational locations were excluded to avoid over-estimation of pseudorandom effects, as described above. Data for the model were from six (86%) of seven GBD superregions, 11 (52%) of 21 GBD regions, and 39 (20%) of 195 countries and territories (figure 1). No data were found for southeast Asia, Oceania, central Asia, the Caribbean, Andean Latin America, central Latin America, or any region of sub-Saharan Africa. Data counts such as these for all diseases are found in the disease-specific summaries in the methods appendix of the GBD 2017 paper on non-fatal disease burden estimation. 42 The estimates of age-standardised prevalence of gastrooesophageal reflux disease for all countries and territories in GBD 2017 are presented in the table.
Mean estimates of age-standardised prevalence of gastro-oesophageal reflux disease for all locations in 2017 ranged from 4408 per 100 000 population in Japan to 14 035 cases per 100 000 population in Saudi Arabia (table). Geographical variation in the age-standardised prevalence of gastro-oesophageal reflux disease in 2017 is shown in figure 2. Standardised for age, gastrooesophageal reflux disease was most prevalent in the USA, Italy, Greece, New Zealand, and several countries in Latin America and the Caribbean (excluding southern Latin America), north Africa and the Middle East, and eastern Europe, at more than 11 000 cases per 100 000 population. Age-standardised prevalence was lowest in high-income Asia Pacific, east Asia, Iceland, France, Denmark, and Switzerland, at less than 7000 cases per 100 000 population. The ratio of age-standardised prevalence among males versus females was 1·0 globally in both 1990 and 2017, ranging from 0·98 to 1·00 across super-regions. Prevalence increased with age, peaking at age 75-79 years overall and for both sexes (18 820 [95% UI 13 770-24 000] cases per 100 000 population for both sexes combined; illustrated for each sex separately in figure 3).
Geographical variation in age-standardised YLD rates reflects variation in prevalence. YLD rates by age also reflect variation in prevalence, with a peak rate at ages 75-79 years globally (appendix p 10). No relationship was seen between age-standardised gastro-oesophageal reflux disease YLD rate and Socio-demographic Index (appendix p 11).

Discussion
We estimated a global increase in total YLDs due to gastro-oesophageal reflux disease between 1990 and 2017, and in YLD rates in populations, but stable YLD rates when standardised to a reference age distribution. This discrepancy between a stable age-standardised YLD rate but rising all-age YLD rate over time reflects higher prevalence in older age groups and the ageing of the global population over time. 53 Age-standardised prevalence of gastro-oesophageal reflux disease is estimated to be highest in the USA, Italy, New Zealand, and countries in Latin America and the Caribbean (excluding southern Latin America), north Africa and the Middle East, and eastern Europe, and lowest in high-income Asia Pacific, east Asia, and some countries in western Europe. In contrast to the global trend and most other regions, high-income North America and high-income Asia Pacific showed increases in the age-standardised YLD rate due to gastro-oesophageal reflux disease between 1990 and 2017. In these regions there could be factors contributing to increasing gastro-oesophageal reflux disease burden beyond just demographic changes. However, additional factors contributing to the changing burden in these two regions and factors associated with spatial variation in gastro-oesophageal reflux disease prevalence were not identified here. The fact that  1990 1991 1992 1993 1994 1995 1996 1997 1998 1999   established risk factors of high BMI, alcohol, and smoking were not predictive in our model raises the question of whether spatial and temporal variation in these results is driven more by measurement error than by underlying epidemiology.
Our results are largely consistent with previous systematic reviews and one meta-analysis of gastrooesophageal reflux disease, reporting prevalence estimates ranging from approximately 10% to 30% in the USA and the Middle East and from 3% to 8% in east Asian countries. 4,38,39 Our regional estimates are similar to the regional pooled estimates in the meta-analysis of Eusebi and colleagues, 4 with higher estimates in the Americas and the Middle East, and lower estimates for Asia, although GBD 2017 estimates were generally lower than Eusebi and colleagues' estimates for near-equivalent geographies. A noteworthy difference is that Eusebi and colleagues estimated very high prevalence in South Asia, 22·1% (95% CI 11·5-35·0), well above the GBD estimate of 7·0% (95% UI 6·2-8·0), but very similar estimates in southeast Asia (7·4% vs 8·1%). 4 These differences are likely to be due to the fact that Eusebi and colleagues did not adjust for variations in study design; more than half of the 106 studies included in Eusebi and colleagues' regional estimates had at least one study design characteristic that would have prompted adjustment in the GBD 2017 modelling approach. 4 These differences are consistent with Eusebi and colleagues' finding of a lower global prevalence estimate based on only a subset of studies that met a more stringent case definition. 4 The systematic review by El-Serag and colleagues 39 also noted higher estimates of gastro-oesophageal reflux disease prevalence for studies published in 1995-2009 compared to studies published before 1995, although it did not report a temporal difference over time for studies published after 1995. Similarly, GBD 2017 prevalence estimates rose between 1990 and 2017, but this rise is largely attenuated with age standardisation, which is not addressed by El-Serag and colleagues. Eusebi and colleagues did not test temporal trends.
Our analysis has several limitations. The first and most important limitation is scarce input data and absence of data for many locations. Prevalence data for modelling gastro-oesophageal reflux disease total 144 location-years, similar to many chronic diseases, such as migraine headache (124 location-years of prevalence data), but substantially lower than better-studied diseases such as diabetes (2340 location-years of prevalence data). 42 Scarce data restrict the precision of estimates for all locations. The absence of data for particular locations requires estimates for those locations to be determined by regional, super-regional, and global estimates. Our estimation of YLDs from prevalence data is also limited by scarce data about the distribution of symptom severity and frequency, and the resulting assumption that these distributions are the same across years, age groups, sexes, and locations. Additional data will be sought in future rounds of GBD, and additional population-based studies of gastro-oesophageal reflux disease prevalence, severity, and symptom frequency should be done, particularly in locations with few or no data.
A second data limitation is that input studies use heterogeneous study designs and are subject to potential biases that are only partially overcome in the DisMod modelling framework. Estimating fixed effects for study design characteristics in successive mixed-effects models essentially corrects for potential study-design biases on the basis of ecological comparisons, and cannot fully adjust for variation in study design if certain designs are preferentially used in some years and locations more than others. In future rounds, we should use pre-modelling adjustments for bias that use internal comparisons of case definitions from validation studies or inter-study comparisons of design features between studies that are well matched in location and time. With additional data and improved premodelling data adjust ments, associations between gastrooesophageal reflux disease prevalence data and established risk factors such as high BMI, obesity, and smoking should be re-evaluated, to see whether they can further strengthen predictions in data-sparse locations. Since data on gastrooesophageal reflux disease are taken primarily from surveys, sometimes with low response rates, that were focused on gastrointestinal symptoms and potentially influenced by commercial interest, future rounds of GBD should seek data from general household surveys with high response rates, and consider adjustments to data from surveys that announce a focus on gastrointestinal symptoms (which might bias participation), have poor response rates, or are commercially sponsored.
A third limitation is that our case definition required an individual to have typical reflux symptoms at least weekly for 12 months. This definition is consistent with a published meta-analysis 4 and similar to expert group recommendations for population-based research on gastro-oesophageal reflux disease, 1 but might miss individuals who have appreciable symptoms over shorter periods of time, those who have atypical symptoms, and those who have asymptomatic mucosal injury and risk of complications. Future rounds of GBD should estimate burden due to these additional presentations of the disease. Conversely, symptom-based definitions might include individuals with similar symptoms not due to reflux of stomach contents, such as those with functional dyspepsia. Differences might exist in the association between symptoms and findings on diagnostic studies by location. Validation studies in representative populations should be done to estimate the predictive value of symptom-based questionnaires compared to more comprehensive and specific case definitions.
Finally, health loss due to conditions for which gastrooesophageal reflux disease is a risk factor (such as oesophageal carcinoma) is accounted for in separate GBD estimates, but the relationship to gastro-oesophageal reflux disease should be made more explicit in future rounds to fully account for the effect of this disease on human health.
Our study has several strengths. We have incorporated more prevalence data sources than previously published systematic reviews and one previous meta-analysis. More importantly, GBD 2017 is, to our knowledge, the first study to apply methods of meta-regression to estimate the prevalence of gastro-oesophageal reflux disease, which offers several advantages. Rather than qualitatively assessing the differences in study design that might explain differences in estimates of epidemiological measures from diverse sources, we have accounted quantitatively for many of these important differences using fixed effects for study-level covariates. Rather than reporting estimates only for age groups, years, and locations for which prevalence data have been collected, we have generated estimates for all age groups, years, and locations, incorporating information from adjacent age groups, years, and locations to calculate the best possible prevalence estimates where no data are available. Although estimates for locations without data are less certain, they provide policy makers and other stakeholders with the best available knowledge about the possible extent of this problem, and a tool by which to gauge the value of further research on this disease relative to expenditures in other areas.
The choice of time period for GBD, 1990-2017, also offers the chance to observe trends in gastro-oesophageal reflux disease epidemiology during a period of increasing obesity prevalence. An association between obesity and gastro-oesophageal reflux disease has been observed in previous studies, [2][3][4] suggesting that gastro-oesophageal reflux disease might rise in the 1990-2017 period. The fact that we did not see a rise in age-standardised prevalence of gastro-oesophageal reflux disease in this period does not undermine the association reported in these studies, which were done at the individual level, and could be due to data or modelling limitations (as discussed previously); it could also imply the existence of other risk factors with a large influence on global gastrooesophageal reflux disease occurrence.
Finally, GBD 2017 is the first study to move beyond measuring gastro-oesophageal reflux disease occurrence to estimating the relative burden that gastro-oesophageal reflux disease imposes in terms of YLDs, facilitating comparison with the burden of other diseases and injuries.
In conclusion, GBD 2017 identifies gastro-oesophageal reflux disease as an important cause of non-fatal health loss, which is increasing because of its association with age and the ageing of the global population. Our estimates also show an increase in prevalence after age standardisation for some locations, but variation in agestandardised prevalence was not associated with known risk factors and might be due to measurement error we could not adjust for with current data and methods. These findings indicate that health-care systems need to be prepared to address the needs of increasing numbers of patients with gastro-oesophageal reflux disease. Further studies are needed to identify useful public health interventions. Given the costs and adverse outcomes associated with symptomatic treatment for gastro-oesophageal reflux disease (such as pulmonary infection and loss of bone-mineral density associated with long-term proton-pump inhibitor use) and the increased risk of oesophageal carcinoma in people with gastro-oesophageal reflux disease, additional large, highquality studies of gastro-oesophageal reflux disease prevalence are needed to verify these findings. Further research is required to identify more modifiable risk factors for gastro-oesophageal reflux disease, and to develop more effective interventions to modify its established risk factors and its relationship to oesophageal carcinoma. 54