Reynolds Risk Score for cardiovascular risk in women

Clinical background

The Reynolds Risk Score was developed to address a concrete weakness in the traditional cardiovascular risk models: too many women end up in an intermediate category in which the decision on preventive treatment, above all a statin, becomes a guess. When the ATP-III model (Adult Treatment Panel III) places a woman at a 5 to 10 per cent 10-year risk, there is little in the model to help the clinician decide whether the risk is closer to 5 or closer to 10, and hence whether observation or treatment is warranted.

The Reynolds Risk Score adds two variables not included in ATP-III: high-sensitivity CRP (hsCRP) as a marker of inflammation, and a parental history of myocardial infarction before the age of 60. The idea is that these factors can reclassify patients in the intermediate categories one way or the other, so that the decision becomes clearer. The score is therefore not primarily a tool for finding high-risk patients, but for identifying with greater confidence those whose risk is so low that intensive prevention is not warranted, and those whose risk is high enough to prompt action.

Calculating the Reynolds Risk Score

The formula rests on a Weibull model with the following expression for the 10-year risk:

$$\text{10-year risk ($$

where $B$ is calculated as:

$$B = 0{.}0799 \times \text{age} + 3{.}137 \times \ln(\text{SBP}) + 0{.}180 \times \ln(\text{hsCRP}) + 1{.}382 \times \ln(\text{total cholesterol}) - 1{.}172 \times \ln(\text{HDL-C})$$ $$+ 0{.}134 \times \text{HbA1c (if diabetes)} + 0{.}818 \times \text{smoker} + 0{.}438 \times \text{family history}$$

The variables are: age in years, systolic blood pressure in mmHg, high-sensitivity CRP in mg/L, total cholesterol and HDL cholesterol in mg/dL, HbA1c in per cent (only if diabetes mellitus is present, otherwise this term is set to zero), current smoking (1 if yes, 0 if no), and a parent with myocardial infarction before the age of 60 (1 if yes, 0 if no).

The derivation cohort consisted of 24,558 initially healthy American women aged 45 years or older, recruited into the Women's Health Study and followed for a median of 10.2 years [1]. Two-thirds ($n = 16{,}400$) were used for model development and one-third ($n = 8158$) for internal validation. The outcome modelled was a composite cardiovascular endpoint: myocardial infarction, ischaemic stroke, coronary revascularisation and cardiovascular death. Model B, the clinically simplified version that constitutes the Reynolds Risk Score, was chosen over a more complex model A by minimising the Bayesian information criterion.

Interpretation in practice

The Reynolds Risk Score generates an estimated absolute 10-year risk as a percentage. In the original derivation and in subsequent validation studies, clinical thresholds of 5, 10 and 20 per cent are used to group risk levels [1, 2]:

The decisive application concerns patients in the intermediate 5 to 10 per cent category, in whom the Reynolds Risk Score reclassified approximately 40 to 50 per cent into a higher or lower category in the derivation cohort, and in whom the reclassification matched the observed event rates better than ATP-III did in every case [1].

Validation and performance

The largest external validation was carried out within the Women's Health Initiative Observational Study (WHI-OS), a multi-ethnic prospective cohort of over 90,000 postmenopausal American women aged 50 to 79 years [2]. In a case-cohort design with 1,722 cardiovascular events (752 myocardial infarctions, 754 ischaemic strokes, 216 cardiovascular deaths) and a subcohort of 1,994 women, the Reynolds model showed adequate calibration for the outcome of a major cardiovascular event, unlike the ATP-III model, which overestimated the risk of coronary heart disease, and the Framingham CVD model, which substantially overestimated the risk of total cardiovascular events.

After recalibration, the Reynolds model showed somewhat better discrimination than ATP-III, with a c-statistic of 0.765 compared with 0.757 ($p = 0{.}03$) [2]. The NRI was 4.9 per cent ($p = 0{.}02$) and the IDI 4.1 per cent ($p < 0{.}0001$) compared with ATP-III. The improvement was seen both in White women (NRI 4.3 per cent, $p = 0{.}04$) and in Black women (NRI 11.4 per cent), although the latter result did not reach statistical significance because of the small number of events. An important observation was that the Reynolds model estimated risk quite differently from the Framingham CVD model: the 10 per cent threshold was crossed by 10 per cent and 41 per cent of the women respectively under the two models, which has direct consequences for statin treatment.

In the MESA study (Multi-Ethnic Study of Atherosclerosis), in which 5,185 participants (53 per cent women) were followed for 10 years with 320 ASCVD events, neither hsCRP nor family history added any measurable improvement in discrimination when added to the recalibrated Pooled Cohort Equation [3]. Only the coronary artery calcium (CAC) score improved the c-statistic modestly (0.74 to 0.76, $p = 0{.}04$). This suggests that the marginal value of hsCRP as an additional variable may be smaller in populations in which the underlying risk model is already more modern.

Limitations

The Reynolds Risk Score was derived and validated exclusively in women. A separate equation with different coefficients exists for men [4], and the women's model must not be applied to male patients.

The model rests on a population of initially healthy American women and has been validated mainly in North American cohorts. The external validation in WHI-OS did show good performance in both Black and White women, but the results for other ethnic groups lacked statistical power [2]. Validation in European populations is largely lacking.

hsCRP is influenced by intercurrent inflammation and the acute phase response. A single raised value may reflect a transient infection rather than chronic vascular inflammation, and should not be interpreted in isolation. Repeat measurement after inflammatory status has normalised is recommended if the value is to form the basis of a risk decision.

The HbA1c term is added only in diabetes and is in practice a proxy for glycaemic control among people with diabetes. In the cohort, people with diabetes were initially under-represented, which increases the uncertainty for this subgroup.

The Reynolds Risk Score today competes with the Pooled Cohort Equations (PCE) from the 2013 ACC/AHA guidelines and with SCORE2/SCORE2-OP from the European guidelines, all of which rest on broader and more contemporary populations. In the MESA data, hsCRP added no further discrimination beyond the PCE [3], which calls into question the incremental value of hsCRP specifically in modern risk models. The value of family history as an additional factor is better established.

References

1. Ridker PM, Buring JE, Rifai N, Cook NR. Development and validation of improved algorithms for the assessment of global cardiovascular risk in women: the Reynolds Risk Score. JAMA. 2007;297(6):611-9. PMID: 17299196
2. Cook NR, Paynter NP, Eaton CB, Manson JE, Martin LW, Robinson JG, Rossouw JE, Wassertheil-Smoller S, Ridker PM. Comparison of the Framingham and Reynolds Risk scores for global cardiovascular risk prediction in the multiethnic Women's Health Initiative. Circulation. 2012;125(14):1748-56. PMID: 22399535
3. Yeboah J, Young R, McClelland RL, Delaney JC, Polonsky TS, Dawood FZ, Blaha MJ, Miedema MD, Sibley CT, Carr JJ, Burke GL, Goff DC Jr, Psaty BM, Greenland P, Herrington DM. Utility of nontraditional risk markers in atherosclerotic cardiovascular disease risk assessment. J Am Coll Cardiol. 2016;67(2):139-147. PMID: 26791059
4. Ridker PM, Paynter NP, Rifai N, Glynn RJ, Cook NR. C-reactive protein and parental history improve global cardiovascular risk prediction: the Reynolds Risk Score for men. Circulation. 2008;118(22):2243-51. PMID: 18997194