Challenges in Estimating the Impact of Vaccination with Sparse Data.

BACKGROUND: The synthetic control model is a powerful tool to quantify the population-level impact of vaccines because it can adjust for trends unrelated to vaccination using a composite of control diseases. Because vaccine impact studies are often conducted using smaller, subnational datasets, we evaluated the performance of synthetic control models with sparse time series data. To obtain more robust estimates of vaccine impacts from noisy time series, we proposed a possible alternative approach, STL+PCA method (seasonal-trend decomposition plus principal component analysis), which first extracts smoothed trends from the control time series and uses them to adjust the outcome. METHODS: Using both the synthetic control and STL+PCA models, we estimated the impact of 10-valent pneumococcal conjugate vaccine on pneumonia hospitalizations among cases <12 months and 80+ years of age during 2004-2014 at the subnational level in Brazil. We compared the performance of these models using simulation analyses. RESULTS: The synthetic control model was able to adjust for trends unrelated to 10-valent pneumococcal conjugate vaccine in larger states but not in smaller states. Simulation analyses showed that the estimates obtained with the synthetic control approach were biased when there were fewer cases, and only 4% of simulations had credible intervals covering the true estimate. In contrast, the STL+PCA analysis had 90% lower bias and had 95% of simulations, with credible intervals covering the true estimate. CONCLUSIONS: Estimates from the synthetic control model might be biased when data are sparse. The STL+PCA model provides more accurate evaluations of vaccine impact in smaller populations.

Impact of Pneumococcal Conjugate Vaccines on Pneumonia Hospitalizations in High- and Low-Income Subpopulations in Brazil.

BACKGROUND: Pneumococcal conjugate vaccines (PCVs) are being used worldwide. A key question is whether the impact of PCVs on pneumonia is similar in low- and high-income populations. However, most low-income countries, where the burden of disease is greatest, lack reliable data that can be used to evaluate the impact. Data from middle-income countries that have both low- and high-income subpopulations can provide a proxy measure for the impact of the vaccine in low-income countries. METHODS: We evaluated the impact of PCV10 on hospitalizations for all-cause pneumonia in Brazil, a middle-income country with localities that span a broad range of human development index (HDI) levels. We used complementary time series and spatiotemporal methods (synthetic controls and hierarchical Bayesian spatial regression) to test whether the decline in pneumonia hospitalizations associated with vaccine introduction varied across the socioeconomic spectrum. RESULTS: We found that the declines in all-cause pneumonia hospitalizations in children and young and middle-aged adults did not vary substantially across low and high HDI subpopulations. Moreover, the estimated declines seen in infants and young adults were associated with higher levels of uptake of the vaccine at a local level. CONCLUSIONS: These results suggest that PCVs have an important impact on hospitalizations for all-cause pneumonia in both low- and high-income populations.

Bayesian Model Averaging with Change Points to Assess the Impact of Vaccination and Public Health Interventions.

BACKGROUND: Pneumococcal conjugate vaccines (PCVs) prevent invasive pneumococcal disease and pneumonia. However, some low-and middle-income countries have yet to introduce PCV into their immunization programs due, in part, to lack of certainty about the potential impact. Assessing PCV benefits is challenging because specific data on pneumococcal disease are often lacking, and it can be difficult to separate the effects of factors other than the vaccine that could also affect pneumococcal disease rates. METHODS: We assess PCV impact by combining Bayesian model averaging with change-point models to estimate the timing and magnitude of vaccine-associated changes, while controlling for seasonality and other covariates. We applied our approach to monthly time series of age-stratified hospitalizations related to pneumococcal infection in children younger 5 years of age in the United States, Brazil, and Chile. RESULTS: Our method accurately detected changes in data in which we knew true and noteworthy changes occurred, i.e., in simulated data and for invasive pneumococcal disease. Moreover, 24 months after the vaccine introduction, we detected reductions of 14%, 9%, and 9% in the United States, Brazil, and Chile, respectively, in all-cause pneumonia (ACP) hospitalizations for age group 0 to <1 years of age. CONCLUSIONS: Our approach provides a flexible and sensitive method to detect changes in disease incidence that occur after the introduction of a vaccine or other intervention, while avoiding biases that exist in current approaches to time-trend analyses.

Estimates of hospitalization attributable to influenza and RSV in the US during 1997-2009, by age and risk status.

BACKGROUND: Estimates of influenza and respiratory syncytial virus (RSV) burden must be periodically updated to inform public health strategies. We estimated seasonal influenza- and RSV-attributable hospitalizations in the US from 1997 to 2009 according to age and risk status (NCT01599390). METHODS: Multiple linear regression modelling was used to attribute hospitalizations to influenza or RSV using virological surveillance and hospitalization data. Hospitalization data were obtained from the US Nationwide Inpatient Sample and virology data were obtained from FluView (Centers for Disease Control and Prevention). Outcomes included any mention of ICD-coded respiratory disease and cardiorespiratory disease diagnoses. We also explored a broader definition of respiratory disease that included mention of relevant respiratory sign/symptoms and viral infection ("respiratory broad"). RESULTS: Applying the respiratory broad outcome, our model attributed ~300,000 and ~200,000 hospitalizations to influenza and RSV, respectively. Influenza A/H3N2 was the predominant cause of influenza-related hospitalizations in most seasons, except in three seasons when influenza B was dominant; likewise, A/H3N2 caused most influenza-related hospitalizations in all age segments, except in children <18 years where the relative contribution of A/H3N2 and B was similar. Most influenza A- and B-related hospitalizations occurred in seniors while approximately one half and one third of all RSV-related events occurred in children 0-4 years and seniors 65+ years, respectively. High-risk status was associated with higher risk of both influenza- and RSV-attributable hospitalizations in adults, but not in children. CONCLUSIONS: Our study assessed the burden of influenza and RSV, information that is important for both cost effectiveness studies and for prioritization of the development of antivirals and vaccines. For seniors, we found that the burdens of influenza and RSV were both substantial. Among children <18 years, about half of all influenza hospitalizations were due to influenza B, most occurring in children without noted risk conditions. RSV hospitalizations among children were confined to those 0-4 years. Our study also demonstrated the importance of the outcome used to estimate hospitalization burden. Our findings highlight the burden of influenza among children regardless of risk status and underscore the prevalence of RSV infections among both young children and older adults.

Estimating the population-level impact of vaccines using synthetic controls.

When a new vaccine is introduced, it is critical to monitor trends in disease rates to ensure that the vaccine is effective and to quantify its impact. However, estimates from observational studies can be confounded by unrelated changes in healthcare utilization, changes in the underlying health of the population, or changes in reporting. Other diseases are often used to detect and adjust for these changes, but choosing an appropriate control disease a priori is a major challenge. The "synthetic controls" (causal impact) method, which was originally developed for website analytics and social sciences, provides an appealing solution. With this approach, potential comparison time series are combined into a composite and are used to generate a counterfactual estimate, which can be compared with the time series of interest after the intervention. We sought to estimate changes in hospitalizations for all-cause pneumonia associated with the introduction of pneumococcal conjugate vaccines (PCVs) in five countries in the Americas. Using synthetic controls, we found a substantial decline in hospitalizations for all-cause pneumonia in infants in all five countries (average of 20%), whereas estimates for young and middle-aged adults varied by country and were potentially influenced by the 2009 influenza pandemic. In contrast to previous reports, we did not detect a decline in all-cause pneumonia in older adults in any country. Synthetic controls promise to increase the accuracy of studies of vaccine impact and to increase comparability of results between populations compared with alternative approaches.

Challenges to estimating vaccine impact using hospitalization data.

Because the real-world impact of new vaccines cannot be known before they are implemented in national programs, post-implementation studies at the population level are critical. Studies based on analysis of hospitalization rates of vaccine-preventable outcomes are typically used for this purpose. However, estimates of vaccine impact based on hospitalization data are particularly prone to confounding, as hospitalization rates are tightly linked to changes in the quality, access and use of the healthcare system, which often occur simultaneously with introduction of new vaccines. Here we illustrate how changes in healthcare delivery coincident with vaccine introduction can influence estimates of vaccine impact, using as an example reductions in infant pneumonia hospitalizations after introduction of the 10-valent pneumococcal conjugate vaccine (PCV10) in Brazil. To this end, we explore the effect of changes in several metrics of quality and access to public healthcare on trends in hospitalization rates before (2008-09) and after (2011-12) PCV10 introduction in 2010. Changes in infant pneumonia hospitalization rates following vaccine introduction were significantly associated with concomitant changes in hospital capacity and the fraction of the population using public hospitals. Importantly, reduction of pneumonia hospitalization rates after PCV10 were also associated with the expansion of outpatient services in several Brazilian states, falling more sharply where primary care coverage and the number of health units offering basic and emergency care increased more. We show that adjustments for unrelated (non-vaccine) trends commonly employed by impact studies, such as use of single control outcomes, are not always sufficient for accurate impact assessment. We discuss several ways to identify and overcome such biases, including sensitivity analyses using different denominators to calculate hospitalizations rates and methods that track changes in the outpatient setting. Employing these practices can improve the accuracy of vaccine impact estimates, particularly in evolving healthcare settings typical of low- and middle-income countries.

Model estimates of the burden of outpatient visits attributable to influenza in the United States.

BACKGROUND: Although many studies have modelled the national burdens of hospitalizations and deaths due to influenza, few studies have considered the outpatient burden. To fill this gap for the United States (US), we applied traditional statistical modelling approaches to time series derived from large medical claims databases held in the private sector. METHODS: We accessed ICD-9-coded office visit data extracted from Truven Health Analytics' MarketScan Commercial database covering about one third of the US population <65 years during 2001-2009, and Medicare Supplemental data covering about one fifth of US seniors 65+ during 2006-2009. We extracted weekly time series of visits due to respiratory diagnoses, otitis media (OM), and urinary tract infections (UTI), a "negative control". We used multiple linear regression modelling to estimate age-specific influenza-related excess in office visits. RESULTS: In the <65 year age group, in the 8 pre-pandemic seasons studied and for the broadest defined respiratory outcome, the model attributed an average of ~14.5 M (Standard deviation [SD] across seasons 3.9 million) office visits to influenza (rate of 5,581/100,000 population). Of these, ~80 % of visits occurred in the 5-17 and 18-49 age group. In school children aged 5-17 year olds and adult 18-64 year age groups the majority of visits were due to influenza B, while A/H3N2 explained most visits in children <5 year olds. The model further attributed ~2.2 M OM visits (SD across seasons 790,000) annually to influenza, of which 86 % of these occurred in children <18 years; this indicates that 6.4 % of all infants <2 years and 4.9 % of all toddlers aged 2-4 years in the US have an influenza-attributable outpatient visit with an OM diagnosis. In seniors 65 years and older, our model attributed ~0.7 M (SD across seasons 351,000) respiratory visits to influenza (rate of 1,887/100,000 population). The model identified no significant excess UTI (negative control) visits in most seasons. CONCLUSIONS: This is to our knowledge a first study of the outpatient burden of influenza in the US in a large database. The model estimated that 10 % of all children <18 years and 4 % of the entire population <65 years seek outpatient care for respiratory illness attributable to influenza annually. TRIAL REGISTRATION: ClinicalTrial.gov, NCT02019732 .

Modelling estimates of the burden of respiratory syncytial virus infection in children in the UK.

OBJECTIVE: The burden of respiratory syncytial virus (RSV) illness is not well characterised in primary care. We estimated the burden of disease attributable to RSV in children in the UK between 1995 and 2009. DESIGN: Time-series regression modelling. SETTING: A multiple linear regression model based on weekly viral surveillance (RSV and influenza, Public Health England), and controlled for non-specific seasonal drivers of disease, estimated the proportion of general practitioner (GP) episodes of care (counted as first visit in a series within 28 days; Clinical Practice Research Datalink, CPRD), hospitalisations (Hospital Episode Statistics, HES) and deaths (Office of National Statistics, ONS) attributable to RSV each season. PARTICIPANTS: Children 0-17 years registered with a GP in CPRD, or with a respiratory disease outcome in the HES or ONS databases. PRIMARY OUTCOME MEASURES: RSV-attributable burden of GP episodes, hospitalisations and deaths due to respiratory disease by age. RSV-attributable burden associated with selected antibiotic prescriptions. RESULTS: RSV-attributable respiratory disease in the UK resulted in an estimated 450 158 GP episodes, 29 160 hospitalisations and 83 deaths per average season in children and adolescents, with the highest proportions in children <6 months of age (14 441/100 000 population, 4184/100 000 and 6/100 000, respectively). In an average season, there were an estimated 125 478 GP episodes for otitis media and 416 133 prescriptions for antibiotics attributable to RSV. More GP episodes, hospitalisations and deaths from respiratory disease were attributable to RSV than to influenza in children under 5 years. CONCLUSIONS: The burden of RSV in children in the UK exceeds that of influenza. RSV in children and adolescents contributes substantially to GP office visits for a diverse range of illnesses, and was associated with an average 416 133 prescribed antibiotic courses per season. Effective antiviral treatments and preventive vaccines are urgently needed for the management of RSV infection in children. TRIAL REGISTRATION NUMBER: NCT01706302.

Modelling estimates of age-specific influenza-related hospitalisation and mortality in the United Kingdom.

BACKGROUND: Influenza is rarely confirmed with laboratory testing and accurate assessment of the overall burden of influenza is difficult. We used statistical modelling methods to generate updated, granular estimates of the number/rate of influenza-attributable hospitalisations and deaths in the United Kingdom. Such data are needed on a continuing basis to inform on cost-benefit analyses of treatment interventions, including vaccination. METHODS: Weekly age specific data on hospital admissions (1997-2009) and on deaths (1997-2009) were obtained from national databases. Virology reports (1996-2009) of influenza and respiratory syncytial virus detections were provided by Public Health England. We used an expanded set of ICD-codes to estimate the burden of illness attributable to influenza which we refer to as 'respiratory disease broadly defined'. These codes were chosen to optimise the balance between sensitivity and specificity. A multiple linear regression model controlled for respiratory syncytial virus circulation, with stratification by age and the presence of comorbid risk status (conditions associated with severe influenza outcomes). RESULTS: In the United Kingdom there were 28,516 hospitalisations and 7163 deaths estimated to be attributable to influenza respiratory disease in a mean season, with marked variability between seasons. The highest incidence rates of influenza-attributable hospitalisations and deaths were observed in adults aged 75+ years (252/100,000 and 131/100,000 population, respectively). Influenza B hospitalisations were highest among 5-17 year olds (12/100,000 population). Of all estimated influenza respiratory deaths in 75+ year olds, 50 % occurred out of hospital, and 25 % in 50-64 year olds. Rates of hospitalisations and death due to influenza-attributable respiratory disease were increased in adults identified as at-risk. CONCLUSIONS: Our study points to a substantial but highly variable seasonal influenza burden in all age groups, particularly affecting 75+ year olds. Effective influenza prevention or early intervention with anti-viral treatment in this age group may substantially impact the disease burden and associated healthcare costs. The high burden of influenza B hospitalisation among 5-17 year olds supports current United Kingdom vaccine policy to extend quadrivalent seasonal influenza vaccination to this age group. TRIAL REGISTRATION: ClinicalTrial.gov, NCT01520935.

Modelling estimates of the burden of Respiratory Syncytial virus infection in adults and the elderly in the United Kingdom.

BACKGROUND: Growing evidence suggests respiratory syncytial virus (RSV) is an important cause of respiratory disease in adults. However, the adult burden remains largely uncharacterized as most RSV studies focus on children, and population-based studies with laboratory-confirmation of infection are difficult to implement. Indirect modelling methods, long used for influenza, can further our understanding of RSV burden by circumventing some limitations of traditional surveillance studies that rely on direct linkage of individual-level exposure and outcome data. METHODS: Multiple linear time-series regression was used to estimate RSV burden in the United Kingdom (UK) between 1995 and 2009 among the total population and adults in terms of general practice (GP) episodes (counted as first consultation ≥28 days following any previous consultation for same diagnosis/diagnostic group), hospitalisations, and deaths for respiratory disease, using data from Public Health England weekly influenza/RSV surveillance, Clinical Practice Research Datalink, Hospital Episode Statistics, and Office of National Statistics. The main outcome considered all ICD-listed respiratory diseases and, for GP episodes, related symptoms. Estimates were adjusted for non-specific seasonal drivers of disease using secular cyclical terms and stratified by age and risk group (according to chronic conditions indicating severe influenza risk as per UK recommendations for influenza vaccination). Trial registration NCT01706302 . Registered 11 October 2012. RESULTS: Among adults aged 18+ years an estimated 487,247 GP episodes, 17,799 hospitalisations, and 8,482 deaths were attributable to RSV per average season. Of these, 175,070 GP episodes (36 %), 14,039 hospitalisations (79 %) and 7,915 deaths (93 %) were in persons aged 65+ years. High- versus low-risk elderly were two-fold more likely to have a RSV-related GP episode or death and four-fold more likely be hospitalised for RSV. In most seasons since 2001, more GP episodes, hospitalisations and deaths were attributable to RSV in adults than to influenza. CONCLUSION: RSV is associated with a substantial disease burden in adults comparable to influenza, with most of the hospitalisation and mortality burden in the elderly. Treatment options and measures to prevent RSV could have a major impact on the burden of RSV respiratory disease in adults, especially the elderly.

Estimates of mortality attributable to influenza and RSV in the United States during 1997-2009 by influenza type or subtype, age, cause of death, and risk status.

BACKGROUND: Influenza and respiratory syncytial virus (RSV) cause substantial mortality from respiratory and other causes in the USA, especially among people aged 65 and older. OBJECTIVES: We estimated the influenza-attributable mortality and RSV-attributable mortality in the USA, stratified by age and risk status, using outcome definitions with different sensitivity and specificity. METHODS: Influenza- and RSV-associated mortality was assessed from October 1997-March 2009 using multiple linear regression modeling on data obtained from designated government repositories. RESULTS: The main outcomes and measures included mortality outcome definitions-pneumonia and influenza, respiratory broad, and cardiorespiratory disease. A seasonal average of 10,682 (2287-16,363), 19,100 (4862-29,245), and 28,169 (6797-42,316) deaths was attributed to influenza for pneumonia and influenza, respiratory broad, and cardiorespiratory outcome definitions, respectively. Corresponding values for RSV were 6211 (4584-8169), 11,300 (8546-14,244), and 17,199 (13,384-21,891), respectively. A/H3N2 accounted for seasonal average of 71% influenza-attributable deaths; influenza B accounted for most (51-95%) deaths during four seasons. Approximately 70% influenza-attributable deaths occurred in individuals ≥75 years, with increasing mortality for influenza A/H3N2 and B, but not A/H1N1. In children aged 0-4 years, an average of 97 deaths was attributed to influenza (A/H3N2 = 49, B = 33, A/H1N1 = 15) and 165 to respiratory broad outcome definition (RSV). Influenza-attributable mortality was 2.94-fold higher in high-risk individuals. CONCLUSIONS: Influenza-attributable mortality was highest in older and high-risk individuals and mortality in children was higher than reported in passive Centers for Disease Control and Prevention surveillance. Influenza B-attributable mortality was higher than A in four of 12 seasons. Our estimates represent an updated assessment of influenza-attributable mortality in the USA.

Effect of 13-valent pneumococcal conjugate vaccine on admissions to hospital 2 years after its introduction in the USA: a time series analysis.

BACKGROUND: In March 2010, 13-valent pneumococcal conjugate vaccine (PCV13) replaced the seven-valent vaccine in the USA. We assessed the effect of PCV13 use on pneumococcus-related admissions to hospital 2 years after the vaccine was introduced, when coverage in children younger than age 5 years had reached 54%. METHODS: We used data from a private inpatient discharge record database. We extracted age-specific data for admissions to hospital per month (July 1-June 30) for all-cause pneumonia, invasive pneumococcal disease, non-invasive pneumococcal pneumonia, and empyema (all coded by International Classification of Diseases 9) for 2005-12. We also extracted data for urinary tract infection and hospital admission for any reason as control outcomes. We assessed incidences of hospital admission before and after the introduction of PCV13 and used a negative binomial multiple regression model to estimate how much of the change in hospital admissions could be attributed to the vaccine. FINDINGS: Our model results showed that PCV13 was associated with significant reductions in hospital admissions for all-cause pneumonia for some children (21% [95% CI 14-28] in children aged <2 years, 17% [7-27] in those aged 2-4 years) and for empyema (50% [95% CI 22-68] for children age <2 years, 46% [21-64] for 2-4 years, and 37% [13-54] for 5-17 years). All-cause pneumonia was significantly reduced in adults aged 18-39 years (12% (6-17) but not for other adult age groups. The vaccine also reduced admissions for invasive pneumococcal pneumonia and non-invasive pneumococcal or lobar pneumonia in children and adults, indicating herd protection, although the reduction was only significant in some age groups. INTERPRETATION: Only 2 years into the US programme, PCV13 significantly reduced residual invasive and non-invasive pneumococcal hospital admissions in children younger than 5 years, as well as in some adult age groups. Our study design captured the total prevented hospital burden (directly and indirectly by herd protection) and also showed a reversal of the PCV7 era increase in paediatric empyema related to strain replacement. FUNDING: Pfizer.

Global mortality estimates for the 2009 Influenza Pandemic from the GLaMOR project: a modeling study.

BACKGROUND: Assessing the mortality impact of the 2009 influenza A H1N1 virus (H1N1pdm09) is essential for optimizing public health responses to future pandemics. The World Health Organization reported 18,631 laboratory-confirmed pandemic deaths, but the total pandemic mortality burden was substantially higher. We estimated the 2009 pandemic mortality burden through statistical modeling of mortality data from multiple countries. METHODS AND FINDINGS: We obtained weekly virology and underlying cause-of-death mortality time series for 2005-2009 for 20 countries covering ∼35% of the world population. We applied a multivariate linear regression model to estimate pandemic respiratory mortality in each collaborating country. We then used these results plus ten country indicators in a multiple imputation model to project the mortality burden in all world countries. Between 123,000 and 203,000 pandemic respiratory deaths were estimated globally for the last 9 mo of 2009. The majority (62%-85%) were attributed to persons under 65 y of age. We observed a striking regional heterogeneity, with almost 20-fold higher mortality in some countries in the Americas than in Europe. The model attributed 148,000-249,000 respiratory deaths to influenza in an average pre-pandemic season, with only 19% in persons <65 y. Limitations include lack of representation of low-income countries among single-country estimates and an inability to study subsequent pandemic waves (2010-2012). CONCLUSIONS: We estimate that 2009 global pandemic respiratory mortality was ∼10-fold higher than the World Health Organization's laboratory-confirmed mortality count. Although the pandemic mortality estimate was similar in magnitude to that of seasonal influenza, a marked shift toward mortality among persons <65 y of age occurred, so that many more life-years were lost. The burden varied greatly among countries, corroborating early reports of far greater pandemic severity in the Americas than in Australia, New Zealand, and Europe. A collaborative network to collect and analyze mortality and hospitalization surveillance data is needed to rapidly establish the severity of future pandemics. Please see later in the article for the Editors' Summary.

Mortality burden of the 2009-10 influenza pandemic in the United States: improving the timeliness of influenza severity estimates using inpatient mortality records.

BACKGROUND: Delays in the release of national vital statistics hinder timely assessment of influenza severity, especially during pandemics. Inpatient mortality records could provide timelier estimates of influenza-associated mortality. METHODS: We compiled weekly age-specific deaths for various causes from US State Inpatient Databases (1990-2010) and national vital statistics (1990-2009). We calculated influenza-attributable excess deaths by season based on Poisson regression models driven by indicators of respiratory virus activity, seasonality, and temporal trends. RESULTS: Extrapolations of excess mortality from inpatient data fell within 11% and 17% of vital statistics estimates for pandemic and seasonal influenza, respectively, with high year-to-year correlation (Spearman's rho = 0.87-0.90, P < 0.001, n = 19). We attribute 14,800 excess respiratory and cardiac deaths (95% CI: 10,000-19,650) to pandemic influenza activity during April 2009-April 2010, 79% of which occurred in people under 65 years. CONCLUSIONS: Modeling inpatient mortality records provides useful estimates of influenza severity in advance of national vital statistics release, capturing both the magnitude and the age distribution of pandemic and epidemic deaths. We provide the first age- and cause-specific estimates of the 2009 pandemic mortality burden using traditional 'excess mortality' methods, confirming the unusual burden of this virus in young populations. Our inpatient-based approach could help monitor mortality trends in other infectious diseases.

Longitudinal analysis of pain in patients with metastatic prostate cancer using natural language processing of medical record text.

OBJECTIVES: To test the feasibility of using text mining to depict meaningfully the experience of pain in patients with metastatic prostate cancer, to identify novel pain phenotypes, and to propose methods for longitudinal visualization of pain status. MATERIALS AND METHODS: Text from 4409 clinical encounters for 33 men enrolled in a 15-year longitudinal clinical/molecular autopsy study of metastatic prostate cancer (Project to ELIminate lethal CANcer) was subjected to natural language processing (NLP) using Unified Medical Language System-based terms. A four-tiered pain scale was developed, and logistic regression analysis identified factors that correlated with experience of severe pain during each month. RESULTS: NLP identified 6387 pain and 13 827 drug mentions in the text. Graphical displays revealed the pain 'landscape' described in the textual records and confirmed dramatically increasing levels of pain in the last years of life in all but two patients, all of whom died from metastatic cancer. Severe pain was associated with receipt of opioids (OR=6.6, p<0.0001) and palliative radiation (OR=3.4, p=0.0002). Surprisingly, no severe or controlled pain was detected in two of 33 subjects' clinical records. Additionally, the NLP algorithm proved generalizable in an evaluation using a separate data source (889 Informatics for Integrating Biology and the Bedside (i2b2) discharge summaries). DISCUSSION: Patterns in the pain experience, undetectable without the use of NLP to mine the longitudinal clinical record, were consistent with clinical expectations, suggesting that meaningful NLP-based pain status monitoring is feasible. Findings in this initial cohort suggest that 'outlier' pain phenotypes useful for probing the molecular basis of cancer pain may exist. LIMITATIONS: The results are limited by a small cohort size and use of proprietary NLP software. CONCLUSIONS: We have established the feasibility of tracking longitudinal patterns of pain by text mining of free text clinical records. These methods may be useful for monitoring pain management and identifying novel cancer phenotypes.