Large Observational Databases Research Articles

A burden of proof study on alcohol consumption and ischemic heart disease

Cohort and case-control data have suggested an association between low to moderate alcohol consumption and decreased risk of ischemic heart disease (IHD), yet results from Mendelian randomization (MR) studies designed to reduce bias have shown either no or a harmful association. Here we conducted an updated systematic review and re-evaluated existing cohort, case-control, and MR data using the burden of proof meta-analytical framework. Cohort and case-control data show low to moderate alcohol consumption is associated with decreased IHD risk – specifically, intake is inversely related to IHD and myocardial infarction morbidity in both sexes and IHD mortality in males – while pooled MR data show no association, confirming that self-reported versus genetically predicted alcohol use data yield conflicting findings about the alcohol-IHD relationship. Our results highlight the need to advance MR methodologies and emulate randomized trials using large observational databases to obtain more definitive answers to this critical public health question.

Impact of random oversampling and random undersampling on the performance of prediction models developed using observational health data

BackgroundThere is currently no consensus on the impact of class imbalance methods on the performance of clinical prediction models. We aimed to empirically investigate the impact of random oversampling and random undersampling, two commonly used class imbalance methods, on the internal and external validation performance of prediction models developed using observational health data.MethodsWe developed and externally validated prediction models for various outcomes of interest within a target population of people with pharmaceutically treated depression across four large observational health databases. We used three different classifiers (lasso logistic regression, random forest, XGBoost) and varied the target imbalance ratio. We evaluated the impact on model performance in terms of discrimination and calibration. Discrimination was assessed using the area under the receiver operating characteristic curve (AUROC) and calibration was assessed using calibration plots.ResultsWe developed and externally validated a total of 1,566 prediction models. On internal and external validation, random oversampling and random undersampling generally did not result in higher AUROCs. Moreover, we found overestimated risks, although this miscalibration could largely be corrected by recalibrating the models towards the imbalance ratios in the original dataset.ConclusionsOverall, we found that random oversampling or random undersampling generally does not improve the internal and external validation performance of prediction models developed in large observational health databases. Based on our findings, we do not recommend applying random oversampling or random undersampling when developing prediction models in large observational health databases.

Impact of CoronaVac on Covid-19 outcomes of elderly adults in a large and socially unequal Brazilian city: A target trial emulation study

BackgroundAlthough CoronaVac was the only Covid-19 vaccine adopted in the first months of the Brazilian vaccination campaign, randomized clinical trials to evaluate its efficacy in elderly adults were limited. In this study, we use routinely collected surveillance and SARS-CoV-2 vaccination and testing data comprising the population of the fifth largest city of Brazil to evaluate the effectiveness of CoronaVac in adults 60+ years old against severe outcomes. MethodsUsing large observational databases on vaccination and surveillance data from the city of Fortaleza, Brazil, we defined a retrospective cohort including 324,302 eligible adults aged ≥60 years to evaluate the effectiveness of the CoronaVac vaccine. The cohort included individuals vaccinated between January 21, 2021, and August 31, 2021, who were matched with unvaccinated persons at the time of rollout following a 1:1 ratio according to baseline covariates of age, sex, and Human Development Index of the neighborhood of residence. Only Covid-19-related severe outcomes were included in the analysis: hospitalization, ICU admission, and death. Vaccine effectiveness for each outcome was calculated by using the risk ratio between the two groups, with the risk obtained by the Kaplan-Meier estimator. ResultsWe obtained 62,643 matched pairs for assessing the effectiveness of the two-dose regimen of CoronaVac. The demographic profile of the matched population was statistically representative of the population of Fortaleza. Using the cumulative incidence as the risk associated with each group, starting at day 14 since the receipt of the second dose, we found an 82.3 % (95 % CI 66.3–93.9) effectiveness against Covid-19-related death, 68.4 % (95 % CI 42.3–86.4) against ICU admission, and 55.8 % (95 % CI 42.7–68.3) against hospital admission. ConclusionsOur results show that, despite critical delays in vaccine delivery and limited evidence in efficacy trial estimates, CoronaVac contributed to preventing deaths and severe morbidity due to Covid-19 in elderly adults.

Emulating a Target Trial Using Primary-Care Electronic Health Records: Sodium-Glucose Cotransporter 2 Inhibitor Medications and Hemoglobin A1c.

Substantial effort has been dedicated to conducting randomized controlled experiments to generate clinical evidence for diabetes treatment. Randomized controlled experiments are the gold standard for establishing cause and effect. However, due to their high cost and time commitment, large observational databases such as those comprised of electronic health record (EHR) data collected in routine primary care may provide an alternative source with which to address such causal objectives. We used a Canadian primary-care data repository housed at the University of Toronto (Toronto, Ontario, Canada) to emulate a randomized experiment. We estimated the effectiveness of sodium-glucose cotransporter 2 inhibitor (SGLT-2i) medications for patients with diabetes using hemoglobin A1c (HbA1c) as a primary outcome and marker for glycemic control from 2018 to 2021. We assumed an intention-to-treat analysis for prescribed treatment, with analyses based on the treatment assigned rather than the treatment eventually received. We defined the causal contrast of interest as the net change in HbA1c (percent) between the group receiving the standard of care versus the group receiving SGLT-2i medication. Using a counterfactual framework, marginal structural models demonstrated a reduction in mean HbA1c level with the initiation of SGLT-2i medications. These findings provided effect sizes similar to those from earlier clinical trials on assessing the effectiveness of SGLT-2i medications.

Estimation of marginal structural models under irregular visits and unmeasured confounder: calibrated inverse probability weights

Clinical information collected in electronic health records (EHRs) is becoming an essential source to emulate randomized experiments. Since patients do not interact with the healthcare system at random, the longitudinal information in large observational databases must account for irregular visits. Moreover, we need to also account for subject-specific unmeasured confounders which may act as a common cause for treatment assignment mechanism (e.g. glucose-lowering medications) while also influencing the outcome (e.g. Hemoglobin A1c). We used the calibration of longitudinal weights to improve the finite sample properties and to account for subject-specific unmeasured confounders. A Monte Carlo simulation study is conducted to evaluate the performance of calibrated inverse probability estimators using time-dependent treatment assignment and irregular visits with subject-specific unmeasured confounders. The simulation study showed that the longitudinal weights with calibrated restrictions improved the finite sample bias when compared to the stabilized weights. The application of the calibrated weights is demonstrated using the exposure of glucose lowering medications and the longitudinal outcome of Hemoglobin A1c. Our results support the effectiveness of glucose lowering medications in reducing Hemoglobin A1c among type II diabetes patients with elevated glycemic index (ge 8.5 %) using stabilized and calibrated weights.

Disturbance and management effects on forest soil organic carbon stocks in the Pacific Northwest.

Carbon (C)-informed forest management requires understanding how disturbance and management influence soil organic carbon (SOC) stocks at scales relevant to landowners and forest policy and management professionals. The continued growth of data sets and publications allows powerful synthesis approaches to be applied to such questions at increasingly fine scales. Here, we report results from a synthesis that used meta-analysis of published studies and two large observational databases to quantify disturbance and management impacts on SOC stocks. We conducted this, the third in a series of ecoregional SOC assessments, for the Pacific Northwest, which comprises ~8% of the land area but ~12% of the U.S. forest sector C sink. At the ecoregional level, our analysis indicated that fundamental patterns of vegetation, climate, and topography are far more important controls on SOC stocks than land use history, disturbance, or management. However, the same patterns suggested that increased warming, drying, wildland fire, and forest regeneration failure pose significant risks to SOC stocks across the region. Detailed meta-analysis results indicated that wildfires diminished SOC stocks throughout the soil profile, while prescribed fire only influenced surface organic materials and harvesting had no significant overall impact on SOC. Independent observational data corroborated the negative influence of fire on SOC derived from meta-analysis, suggested that harvest impacts may vary subregionally with climate or vegetation, and revealed that forests with agricultural uses (e.g., grazing) or legacies (e.g., cultivation) had smaller SOC stocks. We also quantified effects of a range of common forest management practices having either positive (organic amendments, nitrogen [N]-fixing vegetation establishment, inorganic N fertilization) or no overall effects on SOC (other inorganic fertilizers, urea fertilization, competition suppression through herbicides). In order to maximize the management applications of our results, we qualified them with ratings of confidence based on degree of support across approaches. Last, similar to earlier published assessments from other ecoregions, we supplemented our quantitative synthesis results with a literature review to arrive at a concise set of tactics for adapting management operations to site-specific criteria.

Learning patient-level prediction models across multiple healthcare databases: evaluation of ensembles for increasing model transportability

BackgroundPrognostic models that are accurate could help aid medical decision making. Large observational databases often contain temporal medical data for large and diverse populations of patients. It may be possible to learn prognostic models using the large observational data. Often the performance of a prognostic model undesirably worsens when transported to a different database (or into a clinical setting). In this study we investigate different ensemble approaches that combine prognostic models independently developed using different databases (a simple federated learning approach) to determine whether ensembles that combine models developed across databases can improve model transportability (perform better in new data than single database models)?MethodsFor a given prediction question we independently trained five single database models each using a different observational healthcare database. We then developed and investigated numerous ensemble models (fusion, stacking and mixture of experts) that combined the different database models. Performance of each model was investigated via discrimination and calibration using a leave one dataset out technique, i.e., hold out one database to use for validation and use the remaining four datasets for model development. The internal validation of a model developed using the hold out database was calculated and presented as the ‘internal benchmark’ for comparison.ResultsIn this study the fusion ensembles generally outperformed the single database models when transported to a previously unseen database and the performances were more consistent across unseen databases. Stacking ensembles performed poorly in terms of discrimination when the labels in the unseen database were limited. Calibration was consistently poor when both ensembles and single database models were applied to previously unseen databases.ConclusionA simple federated learning approach that implements ensemble techniques to combine models independently developed across different databases for the same prediction question may improve the discriminative performance in new data (new database or clinical setting) but will need to be recalibrated using the new data. This could help medical decision making by improving prognostic model performance.

Logistic regression models for patient-level prediction based on massive observational data: Do we need all data?

ObjectiveProvide guidance on sample size considerations for developing predictive models by empirically establishing the adequate sample size, which balances the competing objectives of improving model performance and reducing model complexity as well as computational requirements. Materials and MethodsWe empirically assess the effect of sample size on prediction performance and model complexity by generating learning curves for 81 prediction problems (23 outcomes predicted in a depression cohort, 58 outcomes predicted in a hypertension cohort) in three large observational health databases, requiring training of 17,248 prediction models. The adequate sample size was defined as the sample size for which the performance of a model equalled the maximum model performance minus a small threshold value. ResultsThe adequate sample size achieves a median reduction of the number of observations of 9.5%, 37.3%, 58.5%, and 78.5% for the thresholds of 0.001, 0.005, 0.01, and 0.02, respectively. The median reduction of the number of predictors in the models was 8.6%, 32.2%, 48.2%, and 68.3% for the thresholds of 0.001, 0.005, 0.01, and 0.02, respectively. DiscussionBased on our results a conservative, yet significant, reduction in sample size and model complexity can be estimated for future prediction work. Though, if a researcher is willing to generate a learning curve a much larger reduction of the model complexity may be possible as suggested by a large outcome-dependent variability. ConclusionOur results suggest that in most cases only a fraction of the available data was sufficient to produce a model close to the performance of one developed on the full data set, but with a substantially reduced model complexity.

SCALPEL3: A scalable open-source library for healthcare claims databases

ObjectiveThis article introduces SCALPEL3 (Scalable Pipeline for Health Data), a scalable open-source framework for studies involving Large Observational Databases (LODs). It focuses on scalable medical concept extraction, easy interactive analysis, and helpers for data flow analysis to accelerate studies performed on LODs. Materials and methodsInspired from web analytics, SCALPEL3 relies on distributed computing, data denormalization and columnar storage. It was compared to the existing SAS-Oracle SNDS infrastructure by performing several queries on a dataset containing a three years-long history of healthcare claims of 13.7 million patients. Results and discussionSCALPEL3 horizontal scalability allows handling large tasks quicker than the existing infrastructure while it has comparable performance when using only a few executors. SCALPEL3 provides a sharp interactive control of data processing through legible code, which helps to build studies with full reproducibility, leading to improved maintainability and audit of studies performed on LODs. ConclusionSCALPEL3 makes studies based on SNDS much easier and more scalable than the existing framework [1]. It is now used at the agency collecting SNDS data, at the French Ministry of Health and soon at the National Health Data Hub in France [2].

Sentimental Analysis on Twitter using Pig and Hive

Data science is the analytic process to explore new prediction and pattern when to process the collected data. Data analysis is done using large sets of databases and due to them we can easily form patterns and then they could be recognized. This will helpful for prediction of new challenges and circumstances. From the perspective of statistics data analysis of large observational databases has very challenges which made it a research area in abroad as well as in India. Different tools are available in market to process and analyze the large set of data for prediction of future trends and due to which knowledgeable decision should be created. Bigdata and hadoop are one of them. In this paper we have collected 5000 above tweets and then we have done pre-processing over it and then done sentimental analysis so as to get negative and positive tweets and then done prediction over it so as to get the people’s sentiments over a particular person.

Screening drugs for bone fracture risk: a nation-wide longitudinal study using the national SNDS claims database

Large observational healthcare databases (LODs) offer the potential to produce inexpensive studies on patients' real-world behaviour and drug use. While carefully designed risk quantification studies using LODs are well established, adverse drug reactions (ADR) screening remains an unsolved challenge. Black-box or fully-automated approaches for screening often fail to tackle the many methodological and statistical issues that can be raised.
 We propose a methodology halfway between black-box screening and carefully tailored studies, by studying many drugs at a time, without precise a priori on the shape and timing of the potential ADRs. We use this methodology to study the dynamics of antidepressants, hypnotics, neuroleptics molecules and fracture risk as a proxy for fall risk in the elderly, using data from the SNDS, a French large healthcare claims database.

Don’t abandon RCTs in IVF. We don’t even understand them

The conclusion of the Human Fertilisation and Embryology Authority that ‘add-on’ therapies in IVF are not supported by high-quality evidence has prompted new questions regarding the role of the randomized controlled trial (RCT) in evaluating infertility treatments. Critics argue that trials are cumbersome tools that provide irrelevant answers. Instead, they argue that greater emphasis should be placed on large observational databases, which can be analysed using powerful algorithms to determine which treatments work and for whom. Although the validity of these arguments rests upon the sciences of statistics and epidemiology, the discussion to date has largely been conducted without reference to these fields. We aim to remedy this omission, by evaluating the arguments against RCTs in IVF from a primarily methodological perspective. We suggest that, while criticism of the status quo is warranted, a retreat from RCTs is more likely to make things worse for patients and clinicians.

Challenges with quality of race and ethnicity data in observational databases.

We sought to assess the quality of race and ethnicity information in observational health databases, including electronic health records (EHRs), and to propose patient self-recording as an improvement strategy. We assessed completeness of race and ethnicity information in large observational health databases in the United States (Healthcare Cost and Utilization Project and Optum Labs), and at a single healthcare system in New York City serving a racially and ethnically diverse population. We compared race and ethnicity data collected via administrative processes with data recorded directly by respondents via paper surveys (National Health and Nutrition Examination Survey and Hospital Consumer Assessment of Healthcare Providers and Systems). Respondent-recorded data were considered the gold standard for the collection of race and ethnicity information. Among the 160 million patients from the Healthcare Cost and Utilization Project and Optum Labs datasets, race or ethnicity was unknown for 25%. Among the 2.4 million patients in the single New York City healthcare system's EHR, race or ethnicity was unknown for 57%. However, when patients directly recorded their race and ethnicity, 86% provided clinically meaningful information, and 66% of patients reported information that was discrepant with the EHR. Race and ethnicity data are critical to support precision medicine initiatives and to determine healthcare disparities; however, the quality of this information in observational databases is concerning. Patient self-recording through the use of patient-facing tools can substantially increase the quality of the information while engaging patients in their health. Patient self-recording may improve the completeness of race and ethnicity information.

Machine learning methods for developing precision treatment rules with observational data

Clinical trials have identified a variety of predictor variables for use in precision treatment protocols, ranging from clinical biomarkers and symptom profiles to self-report measures of various sorts. Although such variables are informative collectively, none has proven sufficiently powerful to guide optimal treatment selection individually. This has prompted growing interest in the development of composite precision treatment rules (PTRs) that are constructed by combining information across a range of predictors. But this work has been hampered by the generally small samples in randomized clinical trials and the use of suboptimal analysis methods to analyze the resulting data. In this paper, we propose to address the sample size problem by: working with large observational electronic medical record databases rather than controlled clinical trials to develop preliminary PTRs; validating these preliminary PTRs in subsequent pragmatic trials; and using ensemble machine learning methods rather than individual algorithms to carry out statistical analyses to develop the PTRs. The major challenges in this proposed approach are that treatment are not randomly assigned in observational databases and that these databases often lack measures of key prescriptive predictors and mental disorder treatment outcomes. We proposed a tiered case-cohort design approach that uses innovative methods for measuring and balancing baseline covariates and estimating PTRs to address these challenges.

Real-World Evidence, Causal Inference, and Machine Learning

The current focus on real world evidence (RWE) is occurring at a time when at least two major trends are converging. First, is the progress made in observational research design and methods over the past decade. Second, the development of numerous large observational healthcare databases around the world is creating repositories of improved data assets to support observational research. ObjectiveThis paper examines the implications of the improvements in observational methods and research design, as well as the growing availability of real world data for the quality of RWE. These developments have been very positive. On the other hand, unstructured data, such as medical notes, and the sparcity of data created by merging multiple data assets are not easily handled by traditional health services research statistical methods. In response, machine learning methods are gaining increased traction as potential tools for analyzing massive, complex datasets. ConclusionsMachine learning methods have traditionally been used for classification and prediction, rather than causal inference. The prediction capabilities of machine learning are valuable by themselves. However, using machine learning for causal inference is still evolving. Machine learning can be used for hypothesis generation, followed by the application of traditional causal methods. But relatively recent developments, such as targeted maximum likelihood methods, are directly integrating machine learning with causal inference.

Abstract 23: Patient Level Prediction Models Developed With Large Observational Databases Outperformed Existing Clinical Prediction Scores For Stratifying Bleeding Risk In Patients With Atrial Fibrillation

Background: Vitamin K antagonists (e.g., warfarin) and direct oral anticoagulants (DOACs; e.g., rivaroxaban, apixaban, dabigatran, edoxaban) are effective therapies for lowering the risk of thrombotic outcomes including stroke in patients with atrial fibrillation (AF). Like all anticoagulants, they are associated with an increased risk of bleeding. Several models (e.g., ATRIA, ORBIT, HAS-BLED, CHADS 2 and CHA 2 DS 2 -VASc) are available to predict the risk of major bleeding but their performance has not been widely evaluated in observational databases. Objective: To develop patient-level bleeding risk prediction models and evaluate their performance compared to existing clinical models of bleeding. Methods: Based on tools available through the Observational Health Data Science and Informatics Collaborative and Observed Medical Outcomes Partnership Common Data Model framework, we developed patient-level prediction (PLP) models to predict the risk of major bleeding events in patients with non-valvular AF who were new users of warfarin or DOACs. A LASSO regularized logistic regression technique including over 100,000 baseline covariates was performed across each of 4 US databases: IBM MarketScan ® (Commercial (CCAE), Medicaid (MDCD), and Medicare Supplemental (MDCR)) and Optum Clinformatics ® Extended Data Mart (Optum). We evaluated the performance of the PLP models compared with the existing models using the Area Under the Curve (AUC), with 75% training and 25% testing sets. The models were then externally validated by applying them to the other three databases. Results: In each database, all the PLP models achieved higher internal validity (AUCs 0.69 - 0.83) compared with the existing clinical models. The highest AUC among the existing clinical models was 0.76 for CHA 2 DS 2 -VASc run on the DOACs new user population in the CCAE database; the comparative AUC for the PLP model was 0.79. The external validation of the PLP models was somewhat lower, the lowest being the new user DOACs models learned on the MDCD database and applied to the other three databases with AUCs between 0.56 and 0.58. This is possibly in part due to differences between patients in MDCD versus other databases (e.g., age, disability, socioeconomic status). The highest performing was the new user warfarin model learned on the Optum database validated on the CCAE database with an AUC of 0.75. Conclusion: A patient level prediction model outperforms many existing clinical bleeding risk models currently in use.

Likelihood-Ratio-Test Methods for Drug Safety Signal Detection from Multiple Clinical Datasets.

Pre- and postmarket drug safety evaluations usually include an integrated summary of results obtained using data from multiple studies related to a drug of interest. This paper proposes three approaches based on the likelihood ratio test (LRT), called the LRT methods, for drug safety signal detection from large observational databases with multiple studies, with focus on identifying signals of adverse events (AEs) from many AEs associated with a particular drug or inversely for signals of drugs associated with a particular AE. The methods discussed include simple pooled LRT method and its variations such as the weighted LRT that incorporates the total drug exposure information by study. The power and type-I error of the LRT methods are evaluated in a simulation study with varying heterogeneity across studies. For illustration purpose, these methods are applied to Proton Pump Inhibitors (PPIs) data with 6 studies for the effect of concomitant use of PPIs in treating patients with osteoporosis and to Lipiodol (a contrast agent) data with 13 studies for evaluating that drug's safety profiles.

Immortal Time Bias in Post-Operative Radiation Studies Utilizing the National Cancer Data Base

Studies employing large observational oncology databases, such as the National Cancer Database (NCDB), are appearing with increasing frequency in the oncology literature. Immortal time bias is a well described and often overlooked type of confounding selection bias that may give the illusion of treatment effectiveness in time to event (survival) analyses via marked alterations in effect sizes (i.e. hazard ratios). ‘Immortal time’ may arise during observational study design when follow-up incorporates a period of time where either death or the measured study outcome cannot occur. Bias in immortal time can occur in analyses designed to evaluate the impact of a specific intervention that calculates overall survival (OS) from time of diagnosis (in contrast to time of treatment), and in study designs that do not account for the time to receipt of adjuvant therapies (e.g. postoperative radiation therapy or PORT). Given the increase in NCDB analyses being used to evaluate the impact of PORT, we sought to examine how often-published studies accounted for time dependent confounders. We performed a PUBMED search using the terms: ‘NCDB,’ ‘National Cancer Data Base,’ ‘postoperative radiation,’ and ‘adjuvant radiation therapy,’ published between 2015-2018. Inclusion for review involved all manuscripts where PORT (with or without chemotherapy) was delivered to any disease site as a comparative arm within the study. This led to review of 42 publications. We specifically assessed how OS was defined and whether any statistical metrics were used to account for potential immortal time bias in the receipt of PORT via landmark analyses or adjusted hazard ratios. Of the 42 publications reviewed, OS was not defined in 8 (19%) studies, was determined from date of diagnosis in 20 (47.6%), from date of treatment start date in 8(19%), or from diagnosis with a one month landmark time in 6 (14.2%). There was no landmark analysis, sequential landmark definition, or an adjusted cox model with a clear time dependent definition in 31 studies (73.8%) when accounting for time to PORT. Amongst the 11 studies that did a landmark analysis, 9 of them (81.8%) had a landmark time with rationale that was not clearly defined. Among the manuscripts analyzed, PORT was concluded to be beneficial in 38 studies (90.4%). Immortal time bias may obscure the effect size (usually via overestimating HR) of PORT as an adjuvant therapy and affect the validity of NCDB hypothesis generating studies. The NCDB has many strengths with its power to generate valid research questions, but its inherent biases require rigorous statistical methods to help validate conclusions. Studies assessing PORT are encouraged to measure OS from treatment start date and perform sequential landmark analyses to help quantify PORT effect size and mitigate this common confounder.

Pattern Identification and Predictions in Data Analysis

Data Mining is an analytic process to explore data in search of consistent patterns and/or systematic relationships between variables, and then to validate the findings by applying the detected patterns to new sets of data. The main target of data mining application is prediction. Predictive data mining is important and it has the most direct business applications in world. The paper briefly explains the process of data mining which consists of three stages: (1) the Initial exploration, (2) Pattern identification with validation, and (3) Deployment (application of the model to new data in order to generate predictions). Data Mining is being done for Patterns and Relationships recognitions in Data analysis, with an emphasis on large Observational data bases. From a statistical perspective Data Mining is viewed as computer automated exploratory data analytical system for large sets of data and it has huge Research challenges in India and abroad as well. Machine learning methods form the core of Data Mining and Decision tree learning. Data mining work is integrated within an existing user environment, including the works that already make use of data warehousing and Online Analytical Processing (OLAP). The paper describes how data mining tools predict future trends and behavior which allows in making proactive knowledge-driven decisions.

Epidemiology of AKI: Utilizing Large Databases to Determine the Burden of AKI.

Large observational databases linking kidney function and other routine patient health data are increasingly being used to study acute kidney injury (AKI). Routine health care data show an apparent rise in the incidence of population AKI and an increase in acute dialysis. Studies also report an excess in mortality and adverse renal outcomes after AKI, although with variation depending on AKI severity, baseline, definition of renal recovery, and the time point during follow-up. However, differences in data capture, AKI awareness, monitoring, recognition, and clinical practice make comparisons between health care settings and periods difficult. In this review, we describe the growing role of large databases in determining the incidence and prognosis of AKI and evaluating initiatives to improve the quality of care in AKI. Using examples, we illustrate this use of routinely collected health data and discuss the strengths, limitations, and implications for researchers and clinicians.