The incidence of hospitalization for endocarditis in the U.S. is 12.7 per 100,000 annually. Even accounting for the fact that multiple hospitalizations might occur during a single episode of IE, the incidence of this illness appears substantially higher than previously reported 
. The age adjusted 2.4% annual increase in hospitalizations between 1998 and 2009 is in keeping with reports of rising rates of sepsis and septicemia during this period 
. However, it represents an acceleration of the trend in IE incidence observed in a review of twelve small population-based studies in Europe and Olmsted County, Minnesota 
. That review found rates (per 100,000) of 1.4–1.8 during the 1970s and 1980s (with one outlier), rising to 2.2–4.9 during the 1990s.
The IE patients we identified resemble those in a 2000–2005 prospective cohort study of 2,781 patients in 25 countries (ICE-PC) where 25% of all cases (and 38% of US patients) had health care-associated infections 
. Several epidemiologic features of our patients – their advanced age, the male predominance, the prevalence of co-morbid illness and cardiac implants, grafts and devices, and the dominant role of S. aureus (and, in 2009, MRSA) – suggest that health care-associated infections were frequent.
Remarkably, co-morbid drug use and HIV, although probably under-reported, fell between 1998 and 2009. Perhaps harm reduction programs stressing needle hygiene, and a shift from intravenous to oral opioid use underlie this decrease 
The rising age of endocarditis patients explained few of our findings. Although death and MI rates were higher among the elderly, time trends for these outcomes did not vary by age. The proportion of IE hospitalizations attributed to S. aureus was slightly higher among the non-elderly (in keeping with findings from the ICE-PC cohort 
) and rose faster. The rise in neurologic complications occurred largely among those under 70, and this group accounted for the entire decline in HIV and substance use-associated cases.
We found marked geographic variation in IE hospitalization rates, with higher rates in poorer communities and in the Northeast. This finding remains unexplained; differences in all-cause hospitalization rates and mentions of substance use and HIV were small. The ICE-PC study also documented marked geographic variation (by continent) in the epidemiology of IE 
The geographic variability we observed may, in part, explain the lower estimates of IE incidence in earlier studies. For instance, a widely-cited U.S. study assessed IE incidence in Olmstead County, Minnesota, a region in the Midwest with a relatively low poverty rate that may not be nationally representative.
Despite changes in recommendations that greatly reduced the use of antibiotic prophylaxis for dental and other procedures around 2007 
, we found no coincident inflection in the time trend of total IE hospitalizations or of the subset caused by non-enterococcal streptococci. These reassuring findings mirror those from England and Wales 
Endocarditis outcomes didn’t improve during the 12-year period we studied - perhaps due to the growing predominance of S. aureus infections. Severe complications increased. At 14.5%, mortality was lower than the 18% rate in the ICE-PC study, which oversampled referral hospitals 
. Our analysis probably undercounts deaths; LOS was only 15 days, and some patients undoubtedly died while completing therapy in convalescent settings. In one study, 12% of IE patients died within one year and a third by four years after discharge 
The rate of valve replacement surgery we observed is far lower than the 48.2% who underwent surgical therapy in the ICE-PC cohort, perhaps because a larger proportion of the ICE-PC patients were cared for in referral centers 
Recent studies report similar neurologic complication rates to those we found 
. In our sample, 9% of IE patients suffered strokes versus 7.8% to 21% in others’ series. We found low rates of meningitis and brain abscess, only 2.1%; prior studies reported meningitis in 5% to 16.5% of patients and brain abscess in 0% to 4% 
. Our finding that CNS infections and stroke were more common with S. aureus infection confirms prior findings 
. Some 
, but not all 
previous studies have, like ours, found an association between neurologic complications and mortality.
Inflation-adjusted costs of care doubled between 1998 and 2009, rising at the same rate as costs for other diagnoses despite the absence of important therapeutic innovations for IE. The cost increase cannot be attributed to IE patients’ increasing age; costs were lowest for the oldest age group.
Several caveats apply to our analysis. While the NIS data offers advantages of sample size, wide geographic coverage, the ability to assess time trends, and lack of referral bias, ascertainment of diagnoses depends upon both diagnostic and coding rigor. The discharge summaries from which the NIS data are compiled serve, in part, as the basis for hospital payment, and are hence subject to regulation and audit. Hospitals are supposed to code all diagnoses treated during the stay. Major diagnoses and procedures (e.g. endocarditis or valve replacement) are unlikely to go uncoded, or to be affected by changes in coding practices during our study period. Conversely, regulations proscribe coding for inactive problems and those not treated during the hospital stay (e.g. a history of IE); hospitals that billed for IE in the absence of active infection would be subject to sanctions.
To explore whether over-diagnosis might explain the high incidence of endocarditis we found, we examined hospitalizations in which the IE diagnosis was not listed first. In such cases the most common first diagnoses were conditions closely associated with IE such as sepsis or “complication of an implanted device”. Similarities in patient demographics, organisms, outcomes and cost between patients whose IE was coded as a first vs. subsequent diagnosis suggests that our case ascertainment strategy accurately reflects physicians’ diagnoses of IE.
However, predisposing factors or complications are probably under-coded, since clinicians and coders might not consider them important to patients’ care or to reimbursement. Hence, the NIS likely under-reports conditions such as intermittent drug use, anatomically abnormal but functionally adequate heart valves, the presence of a pacemaker, or the occurrence of a single seizure or brief bout of ARF.
The NIS does not specify the sequence of events during a hospitalization. Hence, we cannot ascertain whether a valve surgery that occurred during an IE hospitalization represented a cause or a treatment of the infection. However, few cases of early prosthetic valve IE acquired during surgery would come to light prior to discharge.
Our hospital-based analysis may underestimate population IE incidence. Some IE patients may die prior to a hospital admission or (rarely) be diagnosed and treated without being admitted. Similarly, we could not identify hospitalized cases whose IE was never identified and who either died or were cured by treatment for another condition such as osteomyelitis. Our downward adjustment to account for transfers to another hospital probably causes underestimation of the number of unique cases; it assumes that all cases were coded as having IE at the receiving hospital.
On the other hand, as emphasized above, some patients are probably double-counted in our analysis because they underwent more than one hospital admission in the course of a single episode of IE. We would expect such readmissions to be more frequent at hospitals which often complete treatment in the outpatient setting. However, recent IE case series from such institutions indicate that such readmission are uncommon −3% to 5% of all admitted IE patients 
. In any event, it seems unlikely that readmissions of outpatient-treated cases accounts for the sharp upswing in IE admissions we observed. While we’re unaware of any national data on trends in the frequency of outpatient IE treatment, the stable length of IE inpatient stay during the course of our study argues against a large increase in outpatient therapy.
The major caveat is that, unlike case series that include only verified and adjudicated diagnoses, we examined clinically diagnosed IE. Hence, our findings reflect the over-diagnosis (and perhaps under-diagnosis) of IE in actual clinical practice, where physicians must sometimes presumptively treat IE despite inconclusive blood culture and echocardiographic data. In von Reyn’s sentinel study 
, the first to apply the now standard 
, strict definitions of endocarditis, and in the recent prospective ICE-PS study 
, only 85% of clinically diagnosed cases met strict criteria for IE. A 15% downward adjustment would reduce our incidence estimate to 10.8 IE hospitalizations annually per 100,000, a figure that is still about twice previous U.S. estimates. Unless diagnosis accuracy is deteriorating, over-diagnosis could not explain the temporal increase we observed.
IE is much more common than previously believed and continues to increase due to the increasing incidence among individuals 50 and older. Rising numbers of cases of IE are due to S. aureus or associated with cardiac devices and implants, and a falling number is linked to drug use. These trends point to the growing importance of patient safety measures to minimize health care-acquired infection. The marked geographic variation we observed remains unexplained, suggesting the salience of future research on Staphylococcal virulence factors as well as environmental and social factors associated with IE.
Identifying causative organisms
Although the ICD-9-CM includes diagnostic codes identifying specific causative organisms for infectious diseases, only about two-thirds of records carrying a diagnosis of endocarditis included an indication of a causative organism. We considered a patient with endocarditis to have a staphylococcus aureus endocarditis if any of the following diagnostic codes were included among his/her diagnoses: 03811, 03812, 04111, 04112. Other staphylococci were identified using the following diagnostic codes: 0381, 03810, 03819, 0411, 04110, 04119. Streptococcal endocarditis was identified by the presence of diagnostic codes 0380, 0382, 0410, 04100, 04101, 04102, 04103, 04104, 04105, 04109; enterococcal endocarditis by 04104; gram negative endocarditis by 04185, 0384, 03840, 03841, 03842, 03843, 03844, 03849, 0413, 0414, 0415, 0416, 0417; anaerobic endocarditis by 0383, 04182, 04183, 04184; other bacterial endocarditides by 0388, 04189, 04181, 09884, 0830, 0020, 03642; and fungal endocarditis by 1160, 1154, 11281.
Identifying comorbidities, predisposing factors and complications of endocarditis
We identified comorbidities and predisposing factors to endocarditis using the following ICD-9-CM diagnostic codes: For HIV infection: 042, v08, 07953; for opiate, cocaine or amphetamine abuse: 3055, 3056, 3057, 30550, 30560, 30570, 30551, 30561, 30571, 30552, 30562, 30572, 3040, 30400, 30401, 30402, 3042, 30420, 30421, 30422, 3044, 30440, 30441, 30442, 3047, 30470, 30471, 30472, 3048, 30480, 30481, 30482, 3049, 30490, 30491, or 30492; and for patient who were on renal dialysis prior to their endocarditis hospitalization V451 and V4511. We identified pre-existing cardiac abnormalities that predispose to infective endocarditis using the following diagnostic codes: for congenital cardiac anomalies: 7464, 7461, 7466, 74600, 74601, 74602, 74609, 7462, 7463, 7464, 7465, 7465, 7467, 7469, 7468, 74681, 74682, 74683, 74684; for chronic rheumatic heart disease: 3950, 3951, 3952, 3953, 3959, 394, 3940, 3941, 3942, 3949, 396, 3960, 3961, 3962, 3963, 3968, 3969, 397, 3970, 3971, 3979, 398, 3989, 39890, 39891, 3989, 39899; for syphilitic heart disease: 0932, 09320, 09321, 09322, 09323, 09324; and for the presence of a cardiac device, implant, or graft: 9960, 99600, 99601, 99602, 99604, 99661, v433, v4501, v450, v4500, v4502.
We used the following procedure codes to identify patients who underwent cardiac valve replacement during their endocarditis hospitalization: 3521, 3522, 3523, 3524, 3520, 3525, 3526, 3527.
We classified patients as having a possible complication of endocarditis using the following diagnostic codes: for acute renal failure 5845, 5847, 5848, 5849, 5846, 584; for myocardial infarction 41000, 41001, 41002, 41010, 41011, 41012, 41020, 41021, 41022, 41030, 41031, 41032, 41040, 41041, 41042, 41050, 41051, 41052, 41061, 41062, 41070, 41071, 41072, 41080, 41081, 41082, 41090, 41091, 41092; for stroke 430, 431, 4320, 4329, 43301, 43311, 43321, 43331, 43381, 43391, 43401, 43411, 43491; for central nervous system abscess or meningitis 320, 3200, 3201, 3203, 32081, 32082, 32089, 3208, 3209, 11283, 1142, 11501, 11511, 11591, 3240, 3241, 3249; for encephalopathy 3483, 34831, 34830, 34839; and for convulsions 7803,78039 (excludes epilepsy).