Is the use of cholesterol in mortality risk algorithms in clinical guidelines valid? Ten years prospective data from the Norwegian HUNT 2 study

Authors

  • Dag S. Thelle MD PhD,

    Corresponding author
    1. Physician, Professor of Epidemiology, Department of Biostatistics, Institute of Basic Medical Sciences, University of Oslo, Oslo, Norway
      Prof. Dag S. Thelle, Department of Biostatistics, University of Oslo, PO Box 1122, Oslo, N-0317 Norway, E-mail: epidag@gmail.com
    Search for more papers by this author
  • Aage Tverdal PhD,

    1. Statistician, Senior Researcher, Department of Pharmacoepidemiology, Division of Epidemiology, Norwegian Institute of Public Health, University of Oslo, Oslo, Norway
    Search for more papers by this author
  • Randi Selmer PhD

    1. Statistician, Senior Researcher, Division of Epidemiology, Norwegian Institute of Public Health, University of Oslo, Oslo, Norway
    Search for more papers by this author

Prof. Dag S. Thelle, Department of Biostatistics, University of Oslo, PO Box 1122, Oslo, N-0317 Norway, E-mail: epidag@gmail.com

To the editor

We thank Petursson et al. [1] for supplying us [2] with age- and cholesterol-specific mortality. It would be of great interest to see the figures also for cardiovascular disease and ischaemic heart disease mortality as well.

We think that fig. 1 in the article calls for analyses stratified by age. In age 40–59, the incidence of cardiovascular death increases with increasing cholesterol in all strata except for non-smoking women with systolic blood pressure of 140 mmHg or above.

Thus, their conclusion of ‘dangers’ of elevated cholesterol, especially in women, should be more balanced. Furthermore, all confidence intervals in the article, except for total mortality in women, include 1.0, which means that the estimates are compatible with both a positive and an inverse relationship.

We note that departure from the proportional hazards assumptions was evaluated by the Schoenfeld residuals. To which degree the proportional hazards assumptions were violated is not mentioned in the article. It is well known that the association between total cholesterol and vascular mortality decreases with age on the relative scale [3]. The same goes for blood pressure [4]. This suggests that the proportional hazards assumption is violated with age attained as time variable and total cholesterol and/or blood pressure in the model. In addition to the Schoenfeld residuals evaluation, we would like to know the results of the test of the proportional hazards assumption which can readily be obtained in STATA (estat phtest).

We have trouble with the interpretation of fig. 1 in the article. It states that it is 10-year incidence per 1000 person-years. But in the squares in the figure the unit is in per cent (%). Are the 10-year incidences transformed directly to percentages? Or have you used the relationship: Cumulative probability of dying during 10 years =1 − exp(−10-year incidence)? This relationship is valid for all-cause mortality. For cause-specific mortality, such as cardiovascular disease, the competing risk framework should be used. For instance, it is many causes of death that compete for a smoker's life. If fig. 1 expresses incidences, they suggest that the proportional hazards assumption is violated. For non-smoking men, the rate ratio for cholesterol ≥5.5 versus <5.5 mmol L−1 is 0.93 in age 60–74 at systolic blood pressure ≥140 mm/Hg. In age 40–59, the corresponding ratio is 2.50. The same exercise for systolic blood pressure below 140 mmHg gives at ratio of 0.52 in age 60–74 and 3.00 in 40–59. This is an indication that the effect, in relative terms, of cholesterol on cardiovascular disease risk changes with age.

Ancillary