A Revised Framingham Stroke Risk Profile to Reflect Temporal Trends

Abstract

Background

Age-adjusted stroke incidence has decreased over the past 50 years, likely due to changes in the prevalence and impact of various stroke risk factors. An updated version of the Framingham Stroke Risk Profile (FSRP) might better predict current risks in the Framingham Heart Study (FHS) and other cohorts. We compared the accuracy of the standard (Old), and of a revised (New) version of the FSRP in predicting the risk of all-stroke and ischemic stroke, and validated this new FSRP in two external cohorts, the 3 Cities (3C) and REGARDS studies.

Methods

We computed the old FSRP as originally described, and a new model that used the most recent epoch-specific risk factors' prevalence and hazard-ratios for persons ≥ 55 years and for the subsample ≥ 65 years (to match the age range in REGARDS and 3C studies respectively), and compared the efficacy of these models in predicting 5- and 10-year stroke risks.

Results

The new FSRP was a better predictor of current stroke risks in all three samples than the old FSRP (Calibration chi-squares of new/old FSRP: in men 64.0/12.1, 59.4/30.6 and 20.7/12.5; in women 42.5/4.1, 115.4/90.3 and 9.8/6.5 in FHS, REGARDS and 3C, respectively). In the REGARDS, the new FSRP was a better predictor among whites compared to blacks.

Conclusions

A more contemporaneous, new FSRP better predicts current risks in 3 large community samples and could serve as the basis for examining geographic and racial differences in stroke risk and the incremental diagnostic utility of novel stroke risk factors.

Background

The Framingham Stroke Risk Profile (FSRP) was originally described in 1991¹; it integrates the effect of age, sex and baseline measurements of various vascular risk factors: systolic blood pressure (SBP), use of antihypertensive medications (HTNRx), presence or absence of left ventricular hypertrophy on electrocardiography (ECG-LVH), prevalent cardiovascular disease (prevalent CVD), current smoking status, and current or previous atrial fibrillation (AF) and diabetes mellitus (DM) to describe a 10-year probability of incident stroke. Thus, in a 70-year old man with elevated levels of one risk factor, a SBP of 160 mm Hg, the risk could vary from 8% in the absence of 6 other risk factors to 85% in the presence of all 6. An aggregate FSRP score is assigned based on the presence or absence (or levels) of individual stroke risk factors and a hazard-ratio (HR) associated with each factor. These HR were derived utilizing data on baseline stroke risk factors gathered in the late 1960s and 70s and prospective surveillance for incident stroke over the 1970s and 80s. This old FSRP has been validated in other cohorts,² is recommended by the American Heart Association, and is used by clinicians to predict individual risk of patients. It has also been used by many studies as an aggregate measure of vascular burden inimical to the brain³ and as a baseline risk predictor to assess the incremental utility of circulating, genetic or imaging biomarkers thought to improve stroke risk prediction.^4–8

However, several studies including Framingham cohorts' data have demonstrated that age-adjusted incidence of stroke has declined over the past 50 years^9–11 and even in the 3 decades since data for the FSRP were first collated.¹¹ When the FSRP was applied to a French study carried out in the early 2000s, it clearly overestimated risk and this has since also been observed in the REasons for Geographic and Ethnic Differences in Stroke (REGARDS) study.^{12, 13} This overestimation could have been due to several factors including population differences (it has been postulated that genetic and lifestyle differences could reduce stroke risk in France compared to the United States), or because an outdated risk profile was being applied to a more contemporaneous sample. The latter appears likely since stroke rates and prevalence of several stroke risk factors have declined whereas stroke preventive treatments have improved in effectiveness in the past 25 years since the FSRP was first described.^{9, 14–16} We examined this question by comparing the predictive accuracy of the old and an updated new FSRP in 3 settings: the more recent Framingham Heart Study (FHS) data, the REGARDS and the 3C study samples.

We present an updated new FSRP based on risk factor prevalence and hazard ratios observed in the FHS over the past two decades (1990 onwards) and validate it in the two external samples.

Methods

Populations

In the FHS, our study sample comprised the Framingham Original cohort enrolled in 1948 and examined biennially since then and the Framingham Offspring and Spouses of Offspring enrolled in 1971 and reexamined approximately once every four years since then.¹⁷ The sample used in developing the old FSRP (n=5,734, age ≥55) was created by pooling two non-overlapping 10-year periods, with data contributed by persons attending the 9th Original (1964–1968) and 14th Original (1975–1978) cohort examinations. The contemporary epoch (n=5,072, age ≥55) was created by pooling two non-overlapping 10-year periods, with data contributed by persons attending the 21st Original (1988–92) and 4th Offspring (1987–91) examinations and by persons attending the 26th Original (1999–2001) and 7th Offspring (1998–2001) examinations. To be included, participants were required to be stroke-free and between 55–84 years of age at the baseline examination. In FHS, all participants were followed for 10 years or until 2010 and provided informed consent approved by the Institutional Review Board at the Boston University Medical Center.

In REGARDS, the sample was enrolled between 2003 and 2007, and comprised 30,239 participants ≥45 years of age, with 45% men and 55% women, 42% black and 58% white. Participants were drawn from all 48 states in the lower Continental US with oversampling in the stroke buckle and belt. Potential participants were identified using commercially available lists from Genesys, Inc (Daily City, California) and were contacted through a personalized mailing followed by a computer-assisted phone interview during which verbal consent was obtained, and information on demographics and stroke risk factors was recorded. A trained health professional visited the participants at their homes to obtain written informed consent, perform ECG, obtain measures of blood pressure, height, weight and blood and urine sampling. Follow-up phone contact was subsequently made at 6-monthly intervals for surveillance of suspected stroke events, with medical records for suspected events retrieved and adjudicated by a panel of physicians. The study was approved by the Institutional Review Boards at all participating universities, and written informed consent was obtained from all participants.¹⁸ Data on up to 5 years of follow-up were included in this analysis.

In the 3C-study, enrollment took place between 1999 and 2001 among persons selected from the electoral rolls of three French cities (Bordeaux, Dijon and Montpellier) who were at least 65 years old, and were not institutionalized at entry. Follow-up visits occurred biennially. The study protocol was approved by the Ethical Committee of the University Hospital of Kremlin-Bicêtre. Written informed consent and permission to access medical records were obtained from each participant. Participant follow-up data for to up to 10 years were included in this analysis.

Definition of Stroke Risk Factors

In FHS

Data on age, presence and absence of individual stroke risk factors and levels of these risk factors were collated from each baseline examination. Age and SBP were treated as continuous variables while other risk factors were treated as dichotomous variables. SBP was recorded as the mean of 2 physician recorded measurements made on the left arm of the seated subject, using a mercury column sphygmomanometer and a cuff of appropriate width. Baseline CVD was recorded as present if coronary artery disease, congestive heart failure or peripheral vascular disease had been documented in the participant at, or prior to, the baseline examination. We defined baseline DM as a fasting plasma glucose level ≥7 mmol/l, a casual plasma glucose level ≥11.1 mmol/L or the use of insulin or an oral hypoglycemic drug at or before the baseline examination.¹⁹ Current cigarette smoking was defined as smoking in the year prior to the baseline examination. We omitted ECG-LVH as its prevalence is very low in the contemporary FHS sample (1.6%) and these data were not available in the 3C study. Participants were considered to have prevalent AF if this arrhythmia was noted by a Framingham cardiologist on any ECG done at or prior to the baseline examination.²⁰

In REGARDS

Age, sex, use of antihypertensive medications, and smoking were recorded as self-reported in the initial phone interview. The definition of DM was identical to that used in FHS and SBP was also directly recorded as the average of 2 seated measurements obtained by a health professional. Non-stroke events were based only on the history and review of study ECG; thus heart disease was defined as self-report of myocardial infarction, coronary artery bypass surgery, angioplasty or stenting, or ECG evidence of myocardial infarction. Determination of AF was by self-report of a physician diagnosis or ECG evidence at baseline. ECG-LVH was determined as in FHS in two-thirds of the participants. However, in a third of the participants (those initially enrolled) ECG-LVH was estimated using a 7-lead rather than the standard 12 lead ECG.²¹

In 3C

Covariates (age, sex, medication use, current smoking status) were defined in an identical manner as in FHS based on an in-person interview except that three measures of SBP were obtained using a digital electronic tensiometer (OMRON M4) and DM was defined as being present if the participant was on hypoglycemic treatment (insulin or oral blood glucose lowering drugs) or if fasting plasma glucose level at the baseline examination was greater than 1.26 g/dL (7 mmol/L); DM was not diagnosed based on random blood sugars. As in REGARDS a subject was classified with cardiovascular disease history when it was evident on baseline ECG, or if the participant reported a history of myocardial infarction, coronary bypass or angioplasty. In addition, in 3C, as in Framingham, CVD was also considered present if there was a history of vascular surgery for lower limb arteritis. AF was determined as in REGARDS and ECG-LVH data were not extracted.

Definition of Stroke

In FHS

Details of stroke surveillance and the protocol for determining the diagnosis and type of stroke have been published previously.^{1, 22} Stroke was defined as an acute onset focal neurological deficit of vascular etiology, persisting for more than 24 hours, concordant with the World Health Organization (WHO) definition; both ischemic and hemorrhagic strokes were included. Individual stroke subtypes were categorized according to an algorithm based on pre-established diagnostic criteria that include clinical features, imaging studies and other laboratory criteria, noninvasive vascular studies, cardiac evaluations for a source of embolus and, when available, information from autopsy studies.²³ Subgroup analysis was confined to the largest subgroup i.e. ischemic strokes; these were diagnosed if a focal neurological deficit was documented and either the imaging showed no hemorrhage, imaging showed an ischemic infarct that correlated with the clinical deficit, or an ischemic infarct was documented at autopsy. Cranial CT or brain MRI was available in 90% of all strokes included in this study. An endpoint review committee reviewed all available records including examination by a Framingham study neurologist in the acute period (within 72 hours) and at one, six, twelve and 24 months after event, wherever feasible, for all individuals with a suspected stroke or those who died during follow-up. This review committee made final decisions regarding the occurrence and type of stroke.

In REGARDS

Participants are contacted via telephone every six months to ascertain vital status and obtain information on reasons for any hospitalization and this interview specifically asks about the occurrence of stroke, transient ischemic attack (TIA) and stroke symptoms. Medical records, including brain imaging, are pursued if the participant reports seeking medical care for stroke or TIA and/or was hospitalized for stroke symptoms or an unknown reason. Once a medical record is received, a committee of stroke neurologists reviews it to verify that a stroke occurred and to assign an ischemic or hemorrhagic stroke subtype. Stroke was defined either according to WHO criteria (same as FHS) or by as non-focal neurologic symptoms and imaging findings consistent with stroke.

In 3C

At each follow-up examination, information was collected concerning suspicion of stroke occurrence. In participants who reported incident stroke (during interview or by self-administered questionnaire), further medical data were collected from general practitioners, specialists, and hospital records where possible. In case of fatal events, emergency medical services and hospital records were used and if unavailable, family physicians and family members were interviewed. An incident stroke was defined as a new focal neurological deficit of sudden or rapid onset, of presumed vascular origin, that lasted 24 hours or more, or leading to death. An endpoint adjudication committee reviewed source documentation and adjudicated diagnosis of stroke according to the criteria of the WHO.

Statistical analyses

For all the analyses, we considered four samples: the FHS sample used in these analyses (called “new FHS sample”), the FHS sample that the old FSRP had been derived from (called “old FHS sample”), the REGARDS sample and the 3C Study sample. All analyses were run separately by sex. We first examined the crude 5-year and 10-year probability of all stroke and ischemic stroke in each sample. We then examined the distributions of risk factors in each sample. Next we investigated the associations between risk factors and incident stroke in each of the four samples. We related the old FSRP (with ECG-LVH set to 0) and the new FSRP to the outcome of first stroke using Cox models in the most recent two decades of data on the FHS cohorts (new FHS sample), and in the entire follow-up period for the 3C and the REGARDS studies. The old FSRP uses the previously published means of the risk factors, hazard ratios and estimates of incident stroke. The new FSRP updates the means of the risk factors to reflect current prevalence, updates the estimate of incident stroke to reflect current rates and updates the hazard ratios to reflect current associations. We used the same risk factors identified in the original FSRP with the exception of LVH, both because estimation of presence or absence of LVH can vary substantially depending on the criteria used, making it less suitable for incorporation into a risk score²¹ and because we showed that in the new FHS sample, it does not improve the risk score performances in either sex.

In order to take into account differences in age at enrollment between the FHS, REGARDS, and 3C studies, we tested interactions of age (studied as 55–64 vs. 65 and above) with various risk factors and when any interaction was significant, an interaction term was included in the model (only the interaction of age with diabetes was significant).

We compared the various models using a modification of the Hosmer-Lemeshow calibration χ² statistic for Cox-models.^{24, 25} This statistic measures how closely the outcomes predicted by a given model approximate the observed outcomes. The performance of the Hosmer-Lemeshow calibration statistic for the Cox model can be biased if there is substantial censoring, however in the FHS dataset which has a censoring rate of <5% this is not a concern.²⁶ We also compared the various models (the 2 versions of the old FSRP and the new FSRP) on each study using the c statistic, a measure analogous to the area under the curve of a receiver-operating curve (ROC). The c statistic²⁷ estimates discrimination, the ability of each model to correctly distinguish between persons who do and those who do not develop an incident stroke over a 5-year period (REGARDS) and 10-year period (New FHS, 3C) of observation. All analyses were undertaken using SAS® version 9.3 (SAS, Cary, NC).

Results

The mean age-adjusted 10-year stroke probability is 6.9% in the new FSRP sample compared to 9.6% in the old FSRP sample¹ suggesting a decrease of 28% over 20 years, consistent with prior estimates.^{9, 21} Tables 1 and 2 show the incidence of total stroke and ischemic stroke (IS) in each study sample and the number of persons at baseline in each study. In the new FHS, over 10 years of follow-up, there were a total of 277 incident strokes (247 ischemic strokes (IS)), 144 (129 IS) in women. In persons aged 55 and above, the probability of all-stroke over 5 years of follow up did not differ for men between the new FHS sample and REGARDS (2.9% vs 3.1%), whereas it was higher for women in REGARDS (2.2% vs. 2.7%). Similar trends were seen for ischemic strokes.

Table 1.

5- and 10-year risk of all-stroke across cohorts

	MEN
	Old FHS	New FHS	REGARDS	REGARDS White	REGARDS Black	New FHS	3C
Sample size	2114	2291	10808	7245	3563	1198	2949
Age range	55 and +	55 and +	55 and +	55 and +	55 and +	65 and +	65 and +
5 Year Follow-up
- Number of Cases		62	309	200	109	45	75
- Kaplan-Meier risk (5-year probability)		0.029	0.031	0.030	0.034	0.041	0.028
10 Year Follow-up
- Number of Cases	150	133				97	124
- Kaplan-Meier risk (10-year probability)	0.084	0.075				0.113	0.054
	WOMEN
	Old FHS	New FHS	REGARDS	REGARDS White	REGARDS Black	New FHS	3C
Sample size	2979	2781	12883	7015	5868	1580	4652
Age range	55 and +	55 and +	55 and +	55 and +	55 and +	65 and +	65 and +
5 Year Follow-up
- Number of Cases		60	320	155	165	53	69
- Kaplan-Meier risk (5-year probability)		0.022	0.027	0.024	0.031	0.035	0.0167
10 Year Follow-up
- Number of Cases	196	144				126	138
- Kaplan-Meier risk (10-year probability)	0.072	0.065				0.104	0.041

Abbreviations: FHS=Framingham Heart Study, REGARDS=REasons for Geographic And Racial Differences in Stroke Study, 3C=3 Cities Study.

Table 2.

5- and 10-year risk of ischemic stroke across cohorts

	MEN
	New FHS	REGARDS	REGARDS White	REGARDS Black	New FHS	3C
Sample size	2291	10808	7245	3564	1198	2949
Age range	55 and +	55 and +	55 and +	55 and +	65 and +	65 and +
5 Year Follow-up
Number of Cases	55	253	163	90	40	55
Kaplan-Meier risk (5-year probability)	0.025	0.026	0.025	0.028	0.036	0.020
10 Year Follow-up
Number of Cases	118				85	95
Kaplan-Meier risk (10-year probability)	0.066				0.100	0.042
	WOMEN
	New FHS	REGARDS	REGARDS White	REGARDS Black	New FHS	3C
Sample size	2781	12883	7015	5868	1580	4652
Age range	55 and +	55 and +	55 and +	55 and +	65 and +	65 and +
5 Year Follow-up
Number of Cases	54	267	136	131	48	57
Kaplan-Meier risk (5-year probability)	0.020	0.023	0.021	0.025	0.032	0.013
10 Year Follow-up
Number of Cases	129				114	112
Kaplan-Meier risk (10-year probability)	0.058				0.095	0.030

Abbreviations: FHS=Framingham Heart Study, REGARDS=REasons for Geographic And Racial Differences in Stroke Study, 3C=3 Cities Study.

In persons aged 65 years old and above, the probability of stroke over 10 years of follow up is still at least twice as high in the new FHS sample compared to the 3C study, for men and women, and for all-stroke as well as for IS.

Prevalence of the Stroke Risk Factors

Table 3 shows the sex specific prevalence at baseline of stroke risk factors in the different cohorts. In men, between the old and the new FHS samples, there has been a substantial decline in the prevalence of smoking (from 34% to 12%) whereas prevalence of atrial fibrillation, diabetes and antihypertensive medication use have increased, ranging from a 50% increase for diabetes to a 160% increase for antihypertensive treatments. In women, similar trends towards a decreased prevalence of smoking were observed as were increases, but to a lesser extent than in men, for atrial fibrillation, diabetes and antihypertensive medication use. The comparison between the new FHS sample and REGARDS among men aged 55 years old and above, showed a higher prevalence of cardiovascular disease, diabetes and antihypertensive medication use among REGARDS participants. Within REGARDS, prevalence of risk factors differed between white and black male participants; in the latter, a higher prevalence of smoking, diabetes and hypertension was observed. The comparison between the new FHS and REGARDS samples among women aged 55 years and above, also showed a higher prevalence of atrial fibrillation, diabetes and antihypertensive medication use among REGARDS participants. Within REGARDS, prevalence of risk factors differed between white and black female participants; in the latter, a higher prevalence of diabetes and hypertension was observed. Comparison between the new FHS and 3C samples among persons aged 65 years and above, suggests that for most risk factors, in men as in women, prevalence tended to be lower in the 3C study, except for diabetes in persons aged 65 and above which had a similar prevalence across cohorts and mean SBP which was higher on average in 3C males than in FHS males.

Table 3.

Prevalence of stroke risk factors at baseline examination for each cohort

COHORT	Old FHS	New FHS	REGARDS	REGARDS, White,	REGARDS, Black,	New FHS	3C
Age range	55 and +	55 and +	55 and +	55 and +	55 and +	65 and +	65 and +
Men
Number of subjects per sample	2372	2291	10808	7245	3563	1198	2949
Mean baseline age (in years)	65.4	66.8	66.7	67.0	66.0	73.0	73.3
Age ≥ 65 years old (%)		52.3	56.4	58.2	52.7	100	100
Current smoking (%)	33.8	12.4	13.1	10.9	17.8	7.4	8.2
Prevalent cardiovascular disease (%)	22.2	18.0	24.1	26.7	18.7	24.5	11.8
Atrial fibrillation (%)	2.8	7.1	8.8	9.8	6.9	10.8	4.5
Diabetes mellitus	10.6	15.1	22.7	18.1	32.1	17.6	13.3
Diabetes mellitus if <65 years (%)		12.3	21.2	16.5	29.7
Diabetes mellitus if ≥65 years (%)		17.6	23.9	19.3	34.2	17.6	13.3
Antihypertensive treatment (%)	16.1	42.1	49.2	44.0	59.7	51.2	47.1
Systolic blood pressure	139	135	129	128	132	138
SBP in non-treated hypertensives (in mmHg)		132	126	125	129	135	147
SBP in treated hypertensives (in mmHg)		139	132	131	134	142	154
Left ventricular hypertrophy	3.5	1.9	9.4	7.2	14.1	2.2
Women
Number of subjects per sample	3362	2781	12883	7015	5868	1580	4652
Mean baseline age (in years)	66.1	67.9	66.1	66.4	65.7	74.1	73.5
Age ≥ 65 years old (%)		56.8	53.3	54.8	51.4	100	100
Current smoking (%)	26.4	13.8	13.3	12.5	14.2	10.5	4.1
Prevalent cardiovascular disease (%)	14.2	10.1	12.8	12.6	13.0	13.9	2.9
Atrial fibrillation (%)	2.2	3.2	8.5	8.9	8.1	4.8	2.5
Diabetes mellitus	7.9	8.9	21.4	13.4	31.0	9.9	7.6
Diabetes mellitus if <65 years (%)		7.4	20.6	12.2	30.0
Diabetes mellitus if ≥65 years (%)		9.9	22.1	14.4	32.0	9.9	7.6
Antihypertensive treatment (%)	25.0	39.7	54.8	44.4	67.3	49.2	47.1
Systolic blood pressure	143	135	127	124	131	140
SBP in non-treated hypertensives (in mmHg)		129	123	121	127	134	139
SBP in treated hypertensives (in mmHg)		144	131	129	133	146	148
Left ventricular hypertrophy	2.9	1.3	10.5	6.2	15.7	2.2

Abbreviations: FHS=Framingham Heart Study, REGARDS=REasons for Geographic And Racial Differences in Stroke Study, 3C=3 Cities Study, SBP=systolic blood pressure.

Impact of the Stroke Risk Factors

Table 4 shows the sex-specific, multivariable hazard-ratios of the various stroke risk factors on total stroke, among men and women in the old and new FHS samples, REGARDS and 3C studies.

Table 4.

Impact of stroke risk factors at baseline examination on 10-years risks of all-stroke in each cohort

Cohorts	Old FHS		New FHS		REGARDS		REGARDS White		REGARDS Black		3C
	HR	95%CI	HR	95%CI	HR	95%CI	HR	95%CI	HR	95%CI	HR	95%CI
Men
Baseline age (10 years)	1.63	[1.33–1.99]	1.64	[1.14–2.38]	1.72	[1.40–2.11]	1.87	[1.70–2.38]	1.38	[0.93–2.05]	2.14	[1.48–3.11]
Current smoking	1.69	[1.27–2.23]	1.60	[0.97–2.67]	2.00	[1.55–2.59]	1.88	[1.32–2.28]	2.07	[1.40–3.08]	1.41	[0.78–2.57]
Prevalent cardiovascular disease	1.73	[1.68–1.78]	1.57	[1.05–2.35]	1.31	[1.06–1.62]	1.14	[0.88–1.47]	1.74	[1.19–2.55]	1.22	[0.73–2.03]
Atrial fibrillation	1.82	[1.01–3.29]	1.08	[0.61–1.94]	1.62	[1.23–2.14]	1.87	[1.38–2.55]	0.97	−0.50–1.86]	2.47	[1.36–4.50]
Age ≥ 65 years old			1.58	[0.79–3.14]	1.34	[0.92–1.96]	1.51	[0.95–2.40]	1.18	[0.61–2.27]
Diabetes mellitus	1.41	[0.97–2.04]
Diabetes mellitus if <65 years (%)			3.87	[1.97–7.61]	1.94	[1.31–2.87]	2.28	[1.33–3.91]	1.51	[0.85–2.68]
Diabetes mellitus if ≥65 years (%)			1.41	[0.87–2.30]	1.15	[0.89–1.49]	1.17	[0.84–1.62]	1.24	[0.80–1.93]	0.98	[0.59–1.65]
Antihypertensive treatment (%)			2.28	[1.32–3.94]	1.26	[0.99–1.60]	1.41	[1.07–1.86]	0.97	[0.61–1.55]	1.39	[0.80–2.41]
Systolic blood pressure	1.16	[1.10–1.24]
SBP in non-treated hypertensives (in mmHg)			1.31	[1.15–1.50]	1.09	[1.00–1.19]	1.07	[0.96–1.20]	1.12	[0.97–1.30]	1.10	[0.97–1.26]
SBP in treated hypertensives (in mmHg)			1.10	[0.99–1.23]	1.05	[1.97–1.29]	1.00	[0.90–1.10]	1.14	[1.01–1.28]	1.11	[1.00–1.23]
Left ventricular hypertrophy	2.20	[1.26–3.84]
(SBP-110)*(200-SBP)*	1.0002	[1.0000,1.0004]
Women
Baseline age (10 years)	2.01	[1.69–2.40]	2.41	[1.72–3.38]	1.83	[1.49–2.24]	2.32	[1.73–3.09]	1.43	[1.05–1.93]	2.94	[2.06–4.21]
Current smoking	1.72	[1.29–2.29]	1.67	[1.01–2.75]	1.77	[1.35–2.32]	2.21	[1.52–3.22]	1.39	[0.94–2.06]	0.77	[0.24–2.43]
Prevalent cardiovascular disease	1.55	[1.17–2.07]	0.97	[0.61–1.56]	1.52	[1.18–1.94]	1.62	[1.15–2.28]	1.43	[0.99–2.06]	0.78	[0.28–2.10]
Atrial fibrillation	3.06	[1.95–4.80]	3.34	[1.92–5.81]	1.57	[1.17–2.10]	1.86	[1.28–2.70]	1.26	[0.78–2.01]	2.66	[1.34–5.27]
Age ≥ 65 years old			1.49	[0.69–3.22]	1.44	[0.97–2.14]	1.06	[0.61–1.86]	2.00	[1.14–3.51]
Diabetes mellitus	1.75	[1.25–2.45]
Diabetes mellitus if <65 years (%)			2.92	[0.95–8.98]	1.66	[1.07–2.59]	2.03	[0.99–4.16]	1.44	[0.82–2.56]
Diabetes mellitus if ≥65 years (%)			1.07	[0.58–1.96]	1.11	[0.84–1.45]	1.22	[0.80–1.87]	0.98	[0.69–1.40]	1.57	[0.94–2.64]
Antihypertensive treatment (%)			1.14	[0.67–1.93]	1.19	[0.93–1.54	0.91	[0.64–1.27]	1.58	[1.03–2.43]	1.31	[0.82–2.08]
Systolic blood pressure	1.17	[1.12–1.23]										-
SBP in non-treated hypertensives (in mmHg)			1.12	[0.98–1.28]	1.16	[1.06–1.26]	1.15	[1.05–1.27]	1.15	[0.97–1.36]	1.14	[1.00–1.29]
SBP in treated hypertensives (in mmHg)			1.19	[1.08–1.30]	1.18	[1.10–1.26[	1.24	[1.12–1.38]	1.13	[1.04–1.23]	1.12	[1.01–1.23]
Left ventricular hypertrophy	2.24	[1.39–3.60]
(SBP-110)*(200-SBP)*	1.0003	[1.0001,1.0004]

Abbreviations: FHS=Framingham Heart Study, 3C=3 Cities Study, REGARDS=REasons for Geographic And Racial Differences in Stroke Study, HR=hazard ratio, CI=confidence interval, SBP=systolic blood pressure.

^*For those using antihypertensive medication and with 110≤SBP≤200; for all others, 0.

Between the old and new FHS samples, in men, the impact of atrial fibrillation on stroke risk has decreased from a hazard ratio of 1.82 to 1.08. In the new FHS sample, among men, an interaction between age and diabetes was observed, the impact of diabetes on stroke risk being much higher in those aged below 65 years (HR=3.87 vs. 1.41). Comparing the old and new FHS samples, in women, the impact of prevalent cardiovascular disease on stroke risk has decreased from a hazard ratio of 1.55 to 0.97. In the new FHS, in women as in men, an interaction between age and diabetes was noted, the impact of diabetes on stroke risk being much higher in those aged below 65 years (HR=2.92 vs. 1.07). Comparing the new FHS and REGARDS studies, the associations were overall stronger in REGARDS than in new FHS sample, for men as well as for women, especially for smoking and atrial fibrillation; exceptions were the risk factors of hypertension in men and atrial fibrillation and diabetes in women that were more strongly associated with increased stroke risk in FHS than in REGARDS. Within the REGARDS sample patterns of association were quite consistent between white and black participants. Comparisons of risk factor impacts on stroke risk between the new FHS and 3C samples showed that for both men and women, age and SBP were independent predictors of total stroke risk in both cohorts. In 3C, for both men and women, AF was also an independent predictor of total stroke. In the new FHS sample, current smoking, and prevalent diabetes in persons aged 55–64 years old, were independent predictors of future stroke risk in both men and women, while prevalent cardiovascular disease and antihypertensive treatment were independent predictors of future stroke risk in men only, and AF was an independent predictor of future stroke risk in women only. Patterns of association of risk factors with the risk of ischemic stroke were similar to those observed with all stroke, and are reported in Supplemental Table 1.

Stroke Risk Prediction based on old FSRP and new FSRP

Table 5 shows the sex-specific c statistic derived using the old (with ECG-LVH set to 0 for all persons), a version of the old FSRP recalibrated to reflect the prevalences, risks and hazard ratios observed in the sample being studied, and the new FSRP in FHS, REGARDS and 3C studies as well as the calibration χ² statistics comparing observed and predicted stroke cases. Values >20 indicate significant lack of calibration (p<0.01). In the calibration analyses using the old FSRP, chi-square values were above 20 and the FSRP significantly overestimated the expected stroke rates in both men and women in FHS and REGARDS and in men in 3C. On the other hand the new FSRP did not over or underestimate stroke risk in either the FHS (Figure 1) or 3C cohorts (Table 5, Figure 2) but did overestimate stroke risk in REGARDS (Table 5, Figure 3), especially among black participants. As expected, the recalibrated FSRP performed well, but data on prevalence, risk, and hazard ratios required to perform a recalibration would not be available in most clinical settings wherein the FSRP is widely recommended and used; the new FSRP can be used in such clinical settings.

Table 5.

10-year risks of stroke: comparison of old FSRP, the recalibrated old FSRP and the new FSRP in the 3 cohorts

	Men						Women
Calibration Chi-square
Cohorts Age range	FHS 55+	REGARDS 55 and +	REGARDS White 55 and+	REGARDS Black 55 and+	FHS 65+	3C 65+	FHS 55+	REGARDS 55 and +	REGARDS White 55 and+	REGARDS Black 55 and+	FHS 65+	3C 65+
Old FSRP, LVH=0	64.0*	59.4*	33.0*	30.8*	50.8*	20.7‡	42.5*	115.4*	42.6*	90.7*	33.7*	9.8
Old FSRP, LVH=0, recalibrated§	29.2*	20.2‡	10.9	14.6	26.8†	17.1	4.4	33.5*	12.4	35.7*	12.1	8.4
New FSRP	12.1	30.6*	20.4‡	25.6†	16.1	12.5	4.1	90.3*	28.5†	102.4*	9.9	6.5
Discrimination: c statistic [Confidence Interval]
Cohort Age range	FHS 55+	REGARDS 55 and +	REGARDS White 55 and+	REGARDS Black 55 and+	FHS 65+	3C 65+	FHS 55+	REGARDS 55 and +	REGARDS White 55 and+	REGARDS Black 55 and+	FHS 65+	3C 65+
Old FSRP, LVH=0	0.74 [0.70–0.77]	0.67 [0.64–0.70]	0.69 [0.66–0.73]	0.64 [0.58–0.69]	0.65 [0.60–0.70]	0.68 [0.64–0.72]	0.77 [0.74–0.81]	0.71 [0.69–0.74]	0.75 [0.71–0.79]	0.66 [0.63–0.71]	0.69 [0.64–0.73]	0.70 [0.67–0.73]
Old FSRP, LVH=0, recalibrated§	0.74 [0.70–0.77]	0.67 [0.64–0.70]	0.69 [0.66–0.73]	0.63 [0.58–0.69]	0.65 [0.60–0.70]	0.68 [0.64–0.72]	0.77 [0.74–0.81]	0.71 [0.69–0.74]	0.75 [0.71–0.79]	0.66 [0.63–0.71]	0.69 [0.64–0.73]	0.70 [0.67–0.73]
New FSRP	0.74 [0.70–0.78]	0.66 [0.63–0.69]	0.68 [0.64–0.72]	0.62 [0.57–0.69]	0.65 [0.60–0.70]	0.70 [0.65–0.75]	0.78 [0.75–0.82]	0.71 [0.68–0.74]	0.75 [0.71–0.79]	0.66 [0.62–0.70]	0.71 [0.66–0.75]	0.72 [0.68–0.76]

Abbreviations: FSRP=Framingham Stroke Risk Profile, FHS=Framingham Heart Study, 3C=3 Cities Study, REGARDS=REasons for Geographic And Racial Differences in Stroke Study, LVH=left ventricular hypertrophy.

^*P<0.001;

^†P< 0.01;

^‡P<0.05.

^§Using means from current data to calculate M (profile (linear function) at the values of the means for each variable in the FSRP), and the new FSRP Cox model evaluated at those means for baseline survival.

Figure 1. Calibration of the Framingham Stroke Risk Profile in the Framingham Heart Study.

A-F. Calibration of the old and new FSRP in the FHS data in (A-C) women and (D-F) men, age 55+. Blue bars show the predicted 10-year stroke risk based on the FSRP and red bars show the actual stroke incidence observed over the 10-year period. X-axes refer to deciles of predicted risk scores. FSRP=Framingham Stroke Risk Profile, FHS=Framingham Heart Study, ECG=electrocardiogram, LVH=left ventricular hypertrophy.

Figure 2. Calibration of the new Framingham Stroke Risk Profile in the 3 Cities Study.

A-B. Calibration of the new FSRP in the 3C data in (A) women and (B) men, age 65+. Blue bars show the predicted 10-year stroke risk based on the FSRP and red bars show the actual stroke incidence observed over the 10-year period. X-axes refer to deciles of predicted risk scores. FSRP=Framingham Stroke Risk Profile, 3C=3 Cities Study.

Figure 3. Calibration of the new Framingham Stroke Risk Profile in the REasons for Geographic And Racial Differences in Stroke Study.

A-F. Calibration of the new FSRP in the REGARDS data in (A) all women, (B) black women, (C) white women, (D) all men, (E) black men, and (F) white men, age 55+. Blue bars show the predicted 5-year stroke risk based on the FSRP and red bars show the actual stroke incidence observed over the 5-year period. X-axes refer to deciles of predicted risk scores. Abbreviations: FSRP=Framingham Stroke Risk Profile, REGARDS= REasons for Geographic And Racial Differences in Stroke Study.

Supplemental Tables 2 and 3 show respectively 1) the risk factors and risks associated with a composite outcome including all stroke and transient Ischemic attack (TIA) cases over ten years of follow-up in Framingham and 2) c statistic derived using the old (with ECG-LVH set to 0 for all persons) and new FSRP in FHS, as well as the calibration χ² statistics comparing observed and predicted stroke + TIA cases. The patterns of results shown in these analyses are very close to those observed when the outcome is limited to all stroke cases.

Figure 4 provides details on how to calculate stroke risks using the new FSRP.

Figure 4. Calculating stroke risks using the new Framingham Stroke Risk Profile.

A. Detailed instructions for calculating the new FSRP, along with several examples. B. The components needed to calculate the FSRP: beta coefficients, means of the risk factors (prevalence for binary variables), M, and S_b(10), baseline survival at 10 years. C. Baseline survival, S_b(t), for t=1,….,10. Abbreviations: FSRP=Framingham Stroke Risk Profile. SMK=current smoking (yes/no), CVD=prevalent cardiovascular disease (yes/no), AF=atrial fibrillation (yes/no), DM=diabetes mellitus (yes/no).

Discussion

Principal Findings and Significance

In the community-based cohort of FHS participants under prospective surveillance for stroke risk factors and incident stroke since 1948, we propose an updated stroke risk function, a new FSRP. Compared to the previous FSRP, it shows better ability to predict future stroke incidence especially in white people across 3 representative cohorts, two in the US and one in France. This new FSRP better reflects the current lower risks of incident stroke in whites, both among men and women. These lower risks seem to be largely driven by four factors: a lower prevalence of smoking, lower mean SBP levels consequent to greater use of antihypertensive medications, a lower impact of two traditional stroke risk factors, atrial fibrillation and prevalent non-stroke CVD (since persons with baseline stroke were excluded in creating this risk score) and finally the aging of the at-risk population combined with a diminished impact of DM on stroke risk among persons over age 65, as compared to younger adults.

Although, we advocate the adoption of this new FSRP, in lieu of the old FSRP for risk stratification, risk prediction and estimation of aggregate exposure to vascular risk factors, our finding that the new FSRP, while better than the old, still underperformed in the REGARDS sample, emphasizes the need for continued, careful collection of risk factor association data from at-risk populations that are different because of race or an incompletely understood higher or lower stroke risk. In such populations, recalibration of the FSRP using population-specific stroke incidence and risk factor impact can then be undertaken.

Comparison with Prior Studies

Typically, recalibration of a risk profile has been undertaken when comparing geographically or ethnically distinct populations;^28–31 however, since the incidence of stroke has decreased over time we believed it was appropriate to undertake recalibration of the FSRP in the same Framingham cohort over a new temporal epoch. Our findings on trends in the prevalence and impact of various traditional stroke risk factors are largely concordant with previous publications.^{10, 11, 32} Prior data from Framingham, and other population samples, have documented a decreased mean SBP and an increasing use of antihypertensive medication, and we observed similar trends in the current study sample; however, the impact of a 10 mm Hg rise in SBP was not significantly different across epochs. Framingham data have shown that over the past 50 years, the incidence of smoking,³³ peripheral vascular disease³⁴ and coronary artery disease among women has decreased³⁵ and that the incidence of congestive heart failure has decreased in men (but not in women);³⁶ however, perhaps because of the increased survival among persons with these conditions, the overall prevalence of CVD did not vary across epochs in our sample. An increased incidence of DM³⁷ and prevalence of AF^{15, 38} have been previously documented in the FHS and other samples,³⁹ as has increasing use of antihypertensive medication in all persons (including diabetics)⁴⁰ and increasing use of warfarin in persons with AF.⁴¹ Our finding of an overall lower stroke incidence in 3C and overall higher stroke incidence in REGARDS has also been previously described; despite this the new FSRP. The observation that temporal trends in the hazard-ratios associated with several of these stroke risk factors differs from the earlier observations of Whisnant and colleagues that odds-ratios associated with individual stroke risk factors did not differ significantly from 1960 to 1984;⁴² we believe this difference is attributable to the longer period of observation in our current study that includes recent decades.

Strengths and Limitations

The strengths of our study are the community-based sample, the reliable and prospective ascertainment of vascular risk factors, the prospective verification of incident stroke using standard clinical definitions, and the excellent continuity among Framingham stroke neurologists, two of whom (PAW and CSK) adjudicated most stroke events over the 3 epochs. Limitations of these data include the fact that the FHS and 3C samples are predominantly of European descent, and replication in the more racially and ethnically diverse REGARDS suggests that the FSRP need to be validated, and if required recalibrated whenever possible, that is wherever reliable, local epidemiologic data are available. However, the modest discrimination ability of the new FSRP in the REGARDS study should be interpreted with caution considering the short follow-up (5 years) in the REGARDS study. Thus our findings should be replicated once the majority of REGARDS participants have reached 10 years of follow-up, a milestone that will be reached over the next 3 years.

Another concern is that at older ages, the competing risk of mortality might affect these risk estimates, however for the short-term 5-year and 10-year risk predictions we present, starting at ages 55 and 65, it remains appropriate to use models that assume survival to the end of the period.

We chose not to examine the role of ‘newer’ stroke risk factors (such as plasma homocysteine and serum CRP levels) in our analyses, which are focused on the risk factors originally examined in the FSRP, but our findings suggest there is room for improvement in discrimination and adding newer risk factors could help, keeping in mind that any additional parameters selected should be routinely available in a practice or population setting and not too strongly correlated with current risk factors in the FSRP.

Finally, the primary purpose of the current analyses was to update the simple clinical FSRP originally described. We expect this will serve as a more contemporaneous model of the FSRP for use by clinicians. An improved model should also include an exploration of the incremental utility, if any, of novel biomarkers but such a composite clinical- and biomarker-based risk prediction model may be less widely applicable that this purely clinical one.

Significance and Future Directions

Our observations highlight the need to undertake periodic recalibration of the FSRP. Our data emphasize that assessments of the prognostic accuracy of the FSRP in different populations should utilize contemporary measures of risk factor impact in the Framingham sample. It is not surprising that an old FSRP based on HR from an earlier epoch could overestimate contemporary risks in a geographically distant or racially diverse population³¹ just as it does in the current epoch within the Framingham sample itself. Similarly, risk stratification models that assess the incremental prognostic utility of novel biomarkers and clinical risk prediction models, should, in our view, be based on periodically recalibrated hazard-ratios. Finally, the periodic generation of such recalibrated data would serve as a population-based index of the efficacy of our ongoing stroke-related primary prevention efforts and would provide the most up-dated risk prediction tool for clinicians.

Clinical Perspective.

What is new?

• This paper revises the very widely used, simple, clinical, Framingham stroke risk prediction score using new Framingham data gathered after 1991, when the previous version was published.

• It reflects current temporal trends in the prevalence and impact of traditional stroke risk factors and in stroke incidence.

• To examine the relevance and validity of this updated risk prediction tool, we have additionally tested it in a European sample from a country with low cardiovascular risk, the 3 Cities Study in France, and in the most nationally representative, racially diverse United States sample, the REGARDS study, which oversamples regions of highest stroke risk.

Why is this important?

• The Framingham Stroke Risk Profile is the most widely used stroke risk prediction tool.

• It has been recommended by the American Heart Association and used in doctor’ offices to educate patients and plan treatments.

• In research settings, it helps assess the incremental predictive value of genetic, circulating or imaging biomarkers.

• Further, it is a good marker of cumulated vascular brain injury and predicts dementia risk.

• Incidence of stroke, stroke risk factors, and treatments for stroke have changed greatly in the past 25 years.

• The new version retains the simplicity of the original tool but updates it to improve accuracy.