If a headline such as “Stroke Tsunami: 30 000 More Strokes in the Young” pops into your Twitter feed based on a study in this month's issue of JAMA Neurology, should the claim be taken seriously? Is urgent action needed to reverse this trend? While that headline would certainly be eye-catching, the best short answer makes for a lede begging to be buried, “Maybe, but the evidence is pretty cloudy.”

Using data from the Nationwide Inpatient Sample-a large and nationally representative administrative database of hospital discharge abstracts-George et al1 found that ischemic stroke hospitalizations are increasing in adults younger than age 65 years in the United States, with almost 30 000 more stroke hospitalizations in 2012 than in 2003, a relative increase in hospitalization rates by 20% to 40%. Similarly, they found across-the-board increases in vascular risk factor diagnoses in young adults discharged with an ischemic stroke. If these findings represent a true epidemiologic trend, understanding the reasons underlying this trend and seeking to reverse it should be a leading priority of the stroke community. However, it is not yet clear whether such urgent action is needed.

To begin, the magnitude of the problem is easily overstated by hyperbolic headline writers (ie, us; this was gracefully resisted by the authors1). While 30 000 more ischemic stroke hospitalizations occurred in the young in 2012 compared with 2003, a substantial proportion of that increase is due to the about 10% growth in the size of the 18-to-54-year-old US population over that interval.2 Thus, population growth alone should be expected to account for about 15 000 additional strokes per year, leaving a rough population-adjusted increase of about 15 000 per year. If stroke in the young has truly increased by that amount, it would represent a pressing public health priority given the severity and potential preventability of the events, although the magnitude might better be described as a large wave rather than a tsunami.

There is considerable uncertainty, however, about whether this trend represents epidemiologic truth or whether it is capturing changes in the measurement system. While it's conceptually simple to count the number of strokes in a population, identifying true trends requires applying consistent definitions and case-capture strategies over time. In the best of cases, such as carefully controlled community-based cohort studies or stroke surveillance studies, these are major challenges. Yet, when counting strokes in administrative data, such as the Nationwide Inpatient Sample, strokes are counted against a background of changing diagnostic classifications, technologies, coding practices, and financial incentives.3 Given the known changes in stroke terminology and stroke care over the period of the study, it is possible to speculate how these changes may have influenced the accuracy of administrative counts.

Three key measurement issues present possible sources of error in administrative stroke ascertainment over the study period. First, changes to the diagnostic classification system may have increased the number of ischemic strokes. During the study period, the definition of transient ischemic attack (TIA) evolved from a time-based to a tissue-based definition.4 In 2003, a patient with 30 minutes of right hemiparesis and a small subcortical infarct on magnetic resonance imaging (MRI) was diagnosed as having a TIA, but in 2012, the diagnosis would have been an ischemic stroke. When the opportunity for diagnostic reclassification occurred in other contexts, large shifts in assigned diagnoses have been noted. For example, as the definition of sepsis evolved and recognition increased, an apparent increase in sepsis diagnoses with secondary pneumonia diagnoses was almost perfectly counterbalanced by a decrease in primary pneumonia diagnoses.5 While it is possible that this represented changing epidemiology, it is considerably more likely to represent reclassification. Second, the use of MRI for neurologic conditions in general,6 and for stroke specifically,7 increased considerably over the period of the study. Thus, it is possible that previously undiagnosed subtle clinical events are now diagnosed as stroke after MRI.8 While it is somewhat reassuring that the presence of an MRI report does not substantially alter diagnostic classification when accompanied by robust clinical information, it is far from obvious that similar conclusions apply to code-based stroke diagnoses.9 Finally, changes in coding practices may contribute to changes in apparent disease prevalence in administrative data. For example, the proportion of status epilepticus discharges increased in patients with primary nonneurologic diagnoses concomitant with reimbursement changes that increased reimbursement when status epilepticus was present.10 A similar dynamic may be in play for ischemic stroke. Hospitals have a financial incentive to select a stroke diagnosis over a TIA diagnosis and relative importance of this incentive may have increased as the costs of stroke care increased. All 3 of these sources of potential measurement error would most likely lead to overcounting rather than undercounting of strokes.

The measured increase in stroke in the young in the Nationwide Inpatient Sample may be owing to a true epidemiologic trend, overstatement of the trend owing to measurement error, or some combination of the 2 factors. Can we make any inferences about which of these possibilities is most likely? The authors1 found an increase in vascular risk factor prevalence over time, which seems to suggest that the trend is a true epidemiologic phenomenon. However, it is also possible that the increase in risk factors over time reflects more accurate coding as part of a general trend toward increasingly accurate coding practices over time11 or changes in the definitions of risk factors. Thus, just as apparent stroke hospitalization rates may change with changes in the coding system, so too may apparent risk factor prevalence. If there was a true increase in risk factors, particularly hypertension, we would have expected a similar upward trend in intracerebral hemorrhage (ICH) hospitalizations, which was not detected. In addition, the lack of change in ICH hospitalization rates suggests the potential importance of changing diagnostic technology use over the study period. Diagnoses of ICH are made largely independent of MRI and the technology used to diagnose ICH, computed tomography, was already widely disseminated prior to the study period. Last, the possibility of differential reclassification of TIA to ischemic stroke over time is suggested both by the increase in MRI use over this time, the large reduction in inpatient mortality, and the trend toward declining TIA hospitalizations over the same period.12

It is startling that in a country that spends almost 20% of the largest gross domestic product on the planet on health care, we cannot say with confidence whether the fifth leading cause of death in the United States is increasing or decreasing in the young. Yet, that is precisely our state of affairs. The best available epidemiologic data that partially overlap this period diverge, with an increasing incidence trend observed in the Greater Cincinnati and Northern Kentucky Stroke Study13 and a flat or slightly down-sloping trend noted in the Brain Attack Surveillance in Corpus Christi Study.14 If a 15 000-person-per-year increase in incidence in disabling stroke in the young would be a public health emergency, then it seems clear that our surveillance systems need to be sensitive enough to detect such a change. If, ultimately, stroke in the young is truly increasing and many of these strokes were preventable, it would represent a massive failure not only of our health care system, but also our stroke surveillance systems for failing to clearly detect the trend and enable it to be acted on. Developing better surveillance systems, then, should be a priority of the stroke research community. A number of potential solutions exist: collecting more widespread epidemiologic data, harmonizing and building cooperation between existing epidemiological studies, or developing more reliable administrative- or electronic medical record-based surveillance systems. As the richest health care system on the planet, the answer to the question of whether stroke in the young is increasing should never be “Maybe.” The authors of the present study1 should be commended for their effort to help resolve this uncertainty.