Million Hearts: Description of the National Surveillance and Modeling Methodology Used to Monitor the Number of Cardiovascular Events Prevented During 2012–2016

Background This study describes the national surveillance and modeling methodology developed to monitor achievement of the Million Hearts initiative's aim of preventing 1 million acute myocardial infarctions, strokes, and other related cardiovascular events during 2012–2016. Methods and Results We calculate sex‐ and age‐specific cardiovascular event rates (combination of emergency department, hospitalization, and death events) among US adults aged ≥18 from 2006 to 2011 and, based on log‐linear models fitted to the rates, calculate their annual percent change. We describe 2 baseline strategies to be used to compare observed versus expected event totals during 2012–2016: (1) stable baselines assume no rate changes, with modeled 2011 rates held constant through 2016; and (2) trend baselines assume 2006–2011 rate trends will continue, with the annual percent changes applied to the modeled 2011 rates to calculate expected 2012–2016 rates. Events prevented estimates during 2012–2013 were calculated using available data: 115 210 (95% CI, 60 858, 169 562) events were prevented using stable baselines and an excess of 43 934 (95% CI, −14 264, 102 132) events occurred using trend baselines. Women aged ≥75 had the most events prevented (stable, 76 242 [42 067, 110 417]; trend, 39 049 [1901, 76 197]). Men aged 45 to 64 had the greatest number of excess events (stable, 22 912 [95% CI, 855, 44 969]; trend, 38 810 [95% CI, 15 567, 62 053]). Conclusions Around 115 000 events were prevented during the initiative's first 2 years compared with what would have occurred had 2011 rates remained stable. Recent flattening or reversals in some event rate trends were observed supporting intensifying national action to prevent cardiovascular events.

I n the US, cardiovascular disease (CVD) accounts for more than $300 billion in annual costs. 1 A major cause of disability 2 and leading cause of death, CVD accounts for nearly 30% of all deaths-more than 800 000 deaths each year. 3 Many CVD-related events, including emergency department (ED) visits, hospitalizations, and deaths, are avoidable, given that they can be attributed to a lack of healthy lifestyle behaviors, preventive health care, or timely and effective medical care. 1,[4][5][6][7] In recent decades, the United States has experienced continuous declines in cardiovascular event rates. 1,8 However, because of an aging population 9 and the high prevalence of CVD risk factors, 1 increased CVD prevalence and a slowing or reversal in previously declining overall cardiovascular event rate trends have been reported up through 2011 and forecast to continue if additional efforts are not taken. [10][11][12][13] Some have suggested that the United States needs to establish and take action on achieving aggressive targets for CVD risk factor prevention and management and continue to improve adherence to evidence-based treatments to make substantial gains in decreasing the CVD burden. [14][15][16] In response, the US Department of Health and Human Services launched the Million Hearts initiative in 2012 to focus, coordinate, and enhance CVD prevention activities across the public and private sectors to prevent 1 million acute myocardial infarctions (AMIs), strokes, and other related cardiovascular events by 2017. 17 Million Hearts is working to meet this goal by focusing its efforts on the ABCS of clinical care-aspirin, when appropriate, for secondary CVD prevention; blood pressure control; cholesterol management; and smoking cessationand in the community setting by reducing overall tobacco use, reducing sodium consumption, and eliminating artificial trans fat from the diet. These intermediate measures have been tracked since Million Hearts began and will continue to be monitored. 18,19 Population-level improvements in these risk factors have been demonstrated to lead to decreases in cardiovascular events. 14,17,[20][21][22] A national surveillance system to collectively track nonfatal and fatal cardiovascular events in the United States is not available, limiting the ability to monitor the effects of the initiative on event totals. 23 The purpose of our study is to describe how Million Hearts is using a novel method to combine ED, hospitalization, and mortality data sets to overcome this limitation in national cardiovascular event surveillance and assess whether the initiative met the aim of preventing 1 million events. Using this methodology, we describe the burden of Million Hearts events before the initiative (2006-2011) and 2 strategies used to predict the expected number of events that would occur during the initiative (2012-2016) without additional intervention and estimate the number of events potentially prevented. We provide events prevented estimates for 2012 and 2013 using currently available data.

Description of Data Sources
We used encounter-level data from the Agency for Healthcare Research and Quality's Healthcare Cost and Utilization Project's Nationwide Emergency Department Sample 24 (NEDS) to capture ED events; annually, it contains 25 to 30 million unweighted ED visit records from over 950 sampled hospitals. We used the provided NEDS weights to calculate national ED estimates and SDs based on the survey's complex sampling design. We used encounter-level data from the Healthcare Cost and Utilization Project's Nationwide Inpatient Sample 25 to capture hospitalization events during 2006 to 2011; annually, it contains %8 million unweighted hospital stays from %1000 sampled hospitals. Beginning in 2012, the Nationwide Inpatient Sample was redesigned 26 as the National Inpatient Sample (collectively referred to throughout as the NIS), selecting a sample of discharges from all hospitals participating in the Healthcare Cost and Utilization Project's, which represents more than 95% of the US population and provides an almost populationlevel account of inpatient utilization. We used the Agency for Healthcare Research and Quality's NIS trend file 27 to establish the hospitalization trend data for 2006-2011 to adjust for the sampling methodology change between 2011 and 2012.
We used mortality data from the Centers for Disease Control and Prevention's National Vital Statistics System (NVSS) Mortality Multiple Cause of Death Files to calculate national mortality estimates and SEs that account for nonsampling-related errors in the death reporting process. 28 The NVSS collects information reported on all death certificates filed in every US state and the District of Columbia.

Identification of Million Hearts Events
The events case definition was developed based on subjectmatter expert input, literature findings, and past coding practices. Multiple decision points were considered in development of the definition (see Data S1), with the most important criterion for an event type's inclusion being that it could be expected to be prevented by current or planned Million Hearts efforts. This led to a definition that included a broader range of event types (eg, included heart failure) than only AMIs and strokes. Although effective from a communications standpoint, the name "Million Hearts" and the aim "to prevent 1 million heart attacks and strokes" 17 are limited in conveying the full complement of the events targeted and assessed. We defined a Million Hearts event as a CVD-related ED encounter or hospitalization with a primary diagnosis International Classification of Diseases, Ninth Revision, Clinical Modification code 29 (specified in Table S1) or a death with an International Classification of Diseases, Tenth Revision underlying cause of death code 30 (specified in Table S1) among adults aged ≥18 years during 2006-2011. We then applied exclusion criteria to each data set to estimate the number of mutually exclusive acute events that were captured in each data set each year ( Figure 1; additional details provided in Figure S1). This involved excluding ED visit events where the patients died in the ED, were transferred to another hospital, or were admitted to the same hospital and excluding hospitalization events that were reported as elective or where the patient died in the hospital or was transferred to another hospital. Million Hearts event totals were then summed across the data sets and SDs calculated (see Data S1). Events were assigned to 1 of 4 categories to describe their distribution-AMIs; strokes; symptomatic precursor conditions (eg, stable angina pectoris); or other CVD conditions (eg, heart failure). The combined totals represent the events tracked by the initiative.
We characterized temporal trends in event rates during 2006-2011 by fitting log-linear regression models to the sexand age-specific rates, weighted by the inverse of each rate's variance. We used the log-transformed regression coefficients (slopes) of the event rate models to calculate the annual percent change (APC) for each rate and the associated SEs and 95% CIs. 32 We tested the APCs within each sex and age group against the null hypothesis (slope=0); we interpreted models with APCs that were statistically different from 0 as having a change in event rate during the period (P<0.05). We compared APCs across age groups for each sex (referent, adults aged 18-44 years) and in each age group by sex (referent, men) to test for statistically significant differences (P<0.05). All statistical tests were 2-sided and conducted using SAS software (9.3; SAS Institute Inc, Raleigh, NC). This study was exempt from institutional review board review because the data analyzed were from de-identified, public-use data sets made available for public health surveillance.

Development and use of two baseline strategies
Two parallel baseline strategies were developed to assess the association between the implementation of Million Hearts and decreases in event totals during 2012-2016. The first strategy (stable baselines) assumes there will be no change in event rates during 2012-2016 unless additional action is taken. To establish the stable baselines, we held the modeled sex-and age-specific 2011 rates constant during 2012-2016. The second strategy (trend baselines) assumes the event rate trends experienced during 2006-2011 will continue during 2012-2016. To establish the trend baselines, we applied the sex-and age-specific APCs to the modeled 2011 rates to calculate expected rates for 2012-2016 (see Data S1 and Figure S2 for an example). We applied the sex-and agespecific modeled 2011 rates and, for the trends baseline, the sex-and age-specific APCs and their associated SEs, to generate 1000 simulated expected rate estimates for each baseline strategy. We then applied the means and SDs of the simulated rates from both baseline strategies against the US Census population estimates 9 Figure S1.
number of events, including the associated 95% CIs, expected to occur each year according to both strategies. NIS, NEDS, and NVSS data were available for 2012 and 2013, so observed event rates and totals were calculated for those years. To estimate the number of events prevented during the first 2 years of the initiative, we subtracted the mean number of expected events obtained from the simulations from the number of actual events observed each year. These differences were then assessed annually and in combination for both years to see whether they were statistically significantly different than 0 (z-score ≥|1.96|). Events prevented estimates, including the use of revised population estimates, will be updated by Million Hearts as data become available for 2014 to 2016.

Event Rates Preceding the Initiative
Prior to the beginning of the initiative, the age-standardized event rate (per 100 000) among all adults decreased 9.8% from 1296.8 (95% CI, 1265.9, 1327.8) in 2006 to 1170.3 (95% CI, 1144.7, 1195.9) in 2011 (Table 1), with similar relative percent decreases in the age-standardized event rates observed among men (9.8%) and women (9.9%; Figure 2). Every age group among both sexes experienced statistically significant declines in event rates during 2006 to 2011 (P<0.05), except for men aged 45 to 64 years ( Table 2). The annual event rate decreased at a statistically significant greater pace among women aged 65 to 74 years compared with women aged 18 to 44 years (P=0.002); it decreased at a statistically significant slower pace among men aged 45 to 64 years compared with men aged 18 to 44 years (P=0.033).

Event Rates by Age and Sex
Event rates increased considerably with age ( Figure 2). Relative rate differences by sex were most striking among the younger age groups. For example, in 2013, the event rate among men aged 45 to 64 years was 64.7% higher than the rate among like-aged women. However, in the oldest age group (aged ≥75 years), the rate among men was only 5.9% higher than the rate among women. Women aged ≥75 years were the largest contributors to the event totals across the 8 sex/age groups in 2013, accounting for 26.5% of the events (Table S2).  to the overall event count in 2013. Comparing the overall agestandardized event rates in 2006 to 2013, by event category, the greatest relative overall reduction in rates occurred for the symptomatic precursor conditions (24.1% relative reduction); the smallest overall reduction occurred for stroke (0.3% relative reduction; Table 1). Moreover, despite considerable improvement in AMI event rates from 2006 to 2010, there was minimal improvement thereafter. Similarly, there was a slight increase in heart failure event rates in 2013 after experiencing declines during 2006 to 2012, which was driven by increases in the overall age-standardized heart failure ED and mortality rates (increased from 63. 8
Using the stable baseline strategy and comparing the expected versus observed sex-and age-specific data, an  33 and obesity 34 and an aging population, 9 to continue making gains in event prevention. The effect these risk factors have on future event rates if they go unimpeded is uncertain; therefore, use of both strategies was considered a reasonable approach. We would consider attainment of the events prevented goal using either strategy a success. However, achieving the goal using the previous trends strategy would be ideal, given that an additional 820 000 events would be prevented if it was met using this strategy compared with only meeting it using the stable baseline strategy.
Although the events prevented during this period cannot be directly attributed to any single intervention or initiative, including Million Hearts, tracking these values is important because it demonstrates the collective progress the United States is making in reducing the burden of CVD and reveals Table 1. The majority of the improvement in overall event rates that did occur during 2012-2013 happened among those aged ≥65 years, especially among women aged ≥75 years. Continued gradual improvement in event rates, in particular among this age group, was likely driven by increased use of cholesterol-lowering medications, 35 including 3-hydroxy-3-methyl-glutaryl-coenzyme A reductase inhibitors (ie, statins), and moderate increases in blood pressure control. 36 Continued improvement in CVD risk factors among this group will likely lead to the largest number of events prevented during the current initiative, given that they have the most events to prevent. In contrast, little to no improvement in event rates occurred among adults aged 45 to 64 years, with rates actually increasing among men aged 45 to 64. Although smoking prevalence continued to decrease among this age group, 37 they remained at greater risk for being unaware or untreated for conditions like hypertension and hypercholesterolemia and less likely to be using aspirin for secondary CVD prevention compared with older adults. 19,38,39 Moreover, this age cohort was born around 1960 and likely has been the generation most affected by the rise of the obesity epidemic. 40 During their lifetime, the prevalence of obesity has more than doubled among adults, 40 possibly placing them at increased risk for developing CVD resulting from overweight status and its associated risk factors compared with previous generations. Therefore, additional focus on preventing or delaying the development of CVD risk factors, 41 identifying and treating the ABCS,  implementing effective weight management interventions, and calculating and managing lifetime CVD risk, 42 instead of just acute risk, is likely required to decrease event rates among this age group now and better position the United States for improved CVD outcomes in future decades. There are potential limitations to this study. One potential limitation is that the event definition is based on mutually exclusive events, not the number of unique persons affected. Although this methodology tries to achieve mutual exclusivity among event totals, there is potential for survivor bias and over-or undercounting events. Moreover, changes in healthcare utilization practices (eg, use of urgent care services instead of ED services) might affect rates rather than changes in disease burden. However, we have attempted to account for this in several ways. First, we excluded all hospitalizations characterized as elective, because how these cases are managed (eg, within acute care hospitals or in outpatient surgical centers) might change over time. Second, by including ED event data, we address potential changes in practice patterns where cases that were previously hospitalized are now managed in the ED setting, including through observation. Finally, the definition relies on use of administrative data; thus, reductions in the number of "silent" events, which account for upward of 20% of all AMIs and strokes 1,43,44 and are likely affected by the focused attention on the leading risk factors, are not captured in monitoring systems and cannot be accounted for. Despite these limitations, the current definition uses the best available data sources to understand the number of unique Million Heartsrelated fatal and nonfatal events prevented at a national level and across all adult age groups and health insurance coverage types, including people without health insurance. The use of this novel methodology helps address the need for improved comprehensive national cardiovascular event surveillance. 23 The NIS was selected to describe the hospitalization burden because it is a robust data set that can be weighted to produce national estimates. While the NIS was redesigned in 2012, use of the Agency for Healthcare Research and Quality's trend weights alleviates differences attributed to sampling rather than changes in utilization during the observation period and the new design is expected to provide more-precise results. 26 The benefits of using the NEDS for ED  Table S2 includes the sex-and age-specific values for each year and for 2012 and 2013 combined. surveillance has been described previously. 45 Although the NEDS does not contain all of the same hospitals as the NIS sample, it is structured similarly; therefore, it better lends itself to the application of the exclusion criteria used to determine mutually exclusive events compared with data sets that are structured differently from the NIS. The biggest potential limitations with use of the NEDS are that the ability to assess trends is somewhat limited because it only provides data since 2006, and, depending on the year, the weighted national estimates are based on discharge data from only 24 to 31 states. However, NEDS allows considerable flexibility in analyses because of its large sample size.
The NVSS was selected to describe the death burden because it is the national standard for describing national death rates and has been used for tracking purposes in scientific publications 1 and by various national and international projects, including the Healthy People initiative. A potential limitation of using the NVSS, as well as the Healthcare Cost and Utilization Project databases, is that they are nonvalidated administrative databases. Use of administrative claims data for surveillance can be affected by reporting or coding changes and has been found to sometimes overestimate cardiovascular event rates, especially deaths. 46 However, the analysis of these data to assess cardiovascular event rate trends are recommended for national surveillance and, when combined across communities, appears valid given that the overestimation likely will not vary greatly at the macro level and over this short period. [46][47][48] Because of the current lag time in data release for each of the data assets used for event surveillance, methodologies and strategies are being explored to forecast and/or calculate more-timely preliminary estimates. In addition, the United States switched from the International Classification of Diseases, Ninth Revision, Clinical Modification-based claims system to the International Classification of Diseases, Tenth Revision, Clinical Modification-based system in October 2015. Crosswalks from the old to the new coding systems have been developed (Table S4); assessment of the effects of applying these new definitions on event totals is underway.
The United States will experience an %6.8% reduction in the total number of events that would have occurred during 2012-2016 by meeting the 1 million events prevented goal using the stable baseline strategy. Whereas only 12% of this goal was achieved in the first 2 years of the initiative through the prevention of an estimated 115 000 events, we expect that the implementation, and therefore the effects, of Million Hearts' focused strategies on overall event totals will increase over time. However, the recent flattening observed in the overall event rate trend is concerning and supports intensifying national action in implementing community-focused policy and systems changes and evidence-based clinical treatment strategies to meet the aggressive, but achievable, targets of Million Hearts. 19 In particular, the increase in event rates observed among those aged 45 to 64 years should serve as a warning that we need to better address the cardiovascular health of this cohort or a reversal in event rate trends may occur. Use of the novel methodology described in this study will allow for Million Hearts and its partners to understand the progress they are making toward their goal of preventing 1 million AMIs, strokes, and other related cardiovascular events through 2016 and beyond.

Acknowledgments
We acknowledge the HCUP Partner States that voluntarily provide their hospitalization and ED data to the Healthcare Cost and Utilization Project (https://www.hcup-us.ahrq.gov/partners.jsp) and each state that provides mortality data to the National Vital Statistics System, without whom the project and data would not be possible.

Sources of Funding
The Centers for Disease Control and Prevention supported this study. No external funding was used. The findings and conclusions in this report are those of the authors and do not necessarily represent the official position of the Centers for Disease Control and Prevention or the Agency for Healthcare Research and Quality.
Data S1. at what level should they be tracked? Yes, stable angina, other acute ischemic heart disease (e.g., unstable angina), transient ischemic attacks, and hypertension should be included in the definition, because they are conditions that: can cause overt symptoms that would potentially lead patients to seek emergency care; are typically non-fatal, especially when treated early; and are usually highly preventable. 1,2 The first three event types remain in the definition across all settings, because they are acute cardiovascular conditions that are often managed in emergency departments (EDs) and during hospitalizations and are highly preventable by Million Hearts activities. Deaths attributable to these conditions are rare, but they were still included in the definition. Hypertension is discussed further below.

Decision Points When Developing the Million Hearts Event Definition
2) Should hypertension be included and at what level? Overall, Million Hearts activities should positively affect hypertension-related ED, hospitalization, and death events via its blood pressure control activities.
However, hypertension-related ED encounters were excluded from the definition because ED services can often be used for hypertension management and not necessarily to treat acute events. With improvements in health insurance coverage through the enactment of the Patient Protection and Affordable Care Act, greater access to primary care services may alter how hypertension is being managed (e.g., clinic-based versus ED-based) for certain groups and changes in event rates may be more indicative of changes in patient access rather than burden. [3][4][5] Hypertension-related hospitalizations and deaths are included in the definition, because current Million Hearts efforts to improve blood pressure control should prevent these acute events from occurring. 6 3) Should heart failure be included in the definition? Or, should only acute heart failure be included? Heart failure events can be prevented by Million Hearts activities. Specifically, improvements in medication adherence 7-9 , a decreased incidence of heart attacks 10 , and improved hypertension control 10 should prevent heart failure-related ED encounters, hospitalizations, and deaths. Originally, we attempted to only include acute heart failure events in the case definition. However, despite the availability of specific codes to identify acute heart failure ED and hospitalization events (ICD-9-CM: 428.21, 428.23, 428.31, 428.33, 428.41, and 428.43) it was found that changes in coding practices during the observation period demonstrated that these codes might not accurately describe the acuity of the event (data not shown).
Therefore, all heart failure event codes (ICD-9-CM: 428.XX) were included in definition. 4) Should acute myocardial infarctions, strokes, and other highly acute ED events be included? A considerable number of higher acuity events (e.g., AMIs and strokes) occurred in the ED. While the vast majority of patients experiencing these events were later hospitalized or died in the ED or hospital and, therefore, were theoretically captured in one of the other two surveillance systems, multiple cases were treated and then released from the hospital. Because of the acuity of these event types, it is expected all patients should have been admitted, transferred, or died. Therefore, we excluded all of the higher acuity ED cases, because they should be captured in the other data sets; those who were not likely had advance directives, were coded incorrectly, or left against medical advice. 5) Should all stroke deaths be included? We included hospitalization and mortality codes that were acute and directly related to ischemic or hemorrhagic strokes, because it was determined that Million Hearts activities can directly impact those conditions. While there is a link between Million Hearts activities (e.g., improving blood pressure control) and decreasing the burden of other types of strokes (e.g., nontraumatic subarachnoid hemorrhages 11 ), those effects are more distal and may not greatly impact hospitalizations; therefore, strokes other than ischemic or hemorrhagic strokes were only included in the mortality definition.

Calculation of Standard Errors for the Mortality Data and Combination of Estimates Across Datasets
Mortality data represent complete counts of all deaths, therefore, they are not subject to sampling error.
However, these data are subject to non-sampling error resulting from potential errors in the death reporting process, therefore, we calculated and report the estimates and standard errors for the mortality counts per the methodology used by the National Center for Health Statistics. 12 The mortality estimates and standard errors were then combined with the weighted estimates and standard deviations derived from the Nationwide Emergency Department Sample and Nationwide/National Inpatient Sample analyses (using the SAS PROC SURVEYFREQ procedure) for the ED visits and hospitalizations, respectively, to report the combined overall event counts and standard deviations. The choice to use the mortality estimates' standard errors for combination, instead of converting the standard errors to standard deviations, was made for two reasons: 1) mortality estimates are not subject to sampling error, therefore, it was determined inappropriate to attempt to calculate standard deviations; and 2) converting the standard errors to standard deviations using the equation: standard deviation = standard error * sqrt ( population n) would greatly overestimate error in the mortality estimates.

Additional Information about Log-linear Model Development and Use to Establish Stable and Previous Trend Baselines and Interpret Progress
The sex-and age-specific rates from 2006-2011 were each fit with a log-linear regression model. The log-transformed beta obtained from each model was then applied within this equation [exp(beta)-1] to describe the annual percent change (APC) in the rates. Because we only had access to 6 years of historical data there wasn't enough available data to appropriately assess for statistically significant breaks in the rate trend using Joinpoint methods or piecewise linear Poisson models. However, during our preliminary analyses we assessed the rates in Joinpoint, 13,14 and no significant trend breaks were identified (P>.05).
Here is how we established the two baselines (stable and previous trends) for each sex-and age-specific group, using men aged 65-74 years as an example ( Figure S2). To establish their stable baseline rate, we held their modeled 2011 rate (2827.4 per 100000; Table 2   # Includes only acute abdominal aortic dissection and rupture (not asymptomatic or elective admissions for surgery).
** "Abdominal aortic aneurysm without mention of rupture" is included because-although not explicitly mentioned-deaths from abdominal aortic aneurysm almost always occur as the result of rupture. † † ICD-10 mortality codes for embolism/thrombosis (I74) do not specify the cause of the embolism/thrombosis like ICD-9-CM codes do, therefore they were excluded. ‡ ‡ "Peripheral vascular disease, unspecified" is included (I73.9), because it best represents deaths attributable to "peripheral artery disease." § § Only deaths attributable to atherosclerosis or peripheral artery disease were included because Million Hearts activities will directly impact mortality related to these types of cases, but not necessarily emergency department visits or hospitalizations.
|||| Cardiac events were only included if they had a secondary diagnosis/contributing cause of death described using a Million Hearts ICD-9-CM diagnosis or ICD-10 death code.