Intended for healthcare professionals

Letters

Early growth and coronary heart disease in later life

BMJ 2001; 323 doi: https://doi.org/10.1136/bmj.323.7312.572 (Published 08 September 2001) Cite this as: BMJ 2001;323:572

Analysis was flawed

  1. T J Cole, professor of medical statistics (tim.cole{at}ich.ucl.ac.uk),
  2. M Fewtrell, MRC senior clinical scientist,
  3. A Lucas, MRC clinical research professor
  1. Centre for Paediatric Epidemiology and Biostatistics, Institute of Child Health, London WC1N 1EH
  2. MRC Childhood Nutrition Research Centre, Institute of Child Health
  3. MRC Environmental Epidemiology Unit (University of Southampton), Southampton General Hospital, Southampton SO16 6YD
  4. National Public Health Institute, Department of Epidemiology and Health Promotion, Diabetes and Genetic Epidemiology Unit, FIN-00300 Helsinki, Finland

    EDITOR—Eriksson et al concluded that in Finnish men born 60 years ago “low weight gain during infancy is associated with increased risk of coronary heart disease,” yet they did not analyse infant weight gain.1 All their references to infant growth relate to size at 1 year (table 3). Had they applied the key regression models that we have described2 to separate the effects of weight at different ages on later outcome, they would have found that infant weight gain was unrelated to risk of coronary heart disease.

    In their simultaneous analysis the hazard ratios for birth weight and weight at 1 year were similar and less than 1, showing that greater weight during infancy is protective. Weight gain is weight at 1 year less weight at birth, so if weight gain were protective it would appear as a protective effect of weight at 1 year and a relatively deleterious effect of weight at birth.2 But the two effects were equally protective, so weight gain in infancy (strictly, upwards centile crossing) is unrelated to later coronary heart disease.

    The hazard ratios for weight at birth and at 1 year can be rearranged as hazard ratios for mean weight and weight gain. The hazard ratio for weight gain is equal approximately to the square root of the ratio of the hazard ratios at 1 year and at birth—that is, Graphic. This is similar to the birthweight hazard ratio, which was not significant (95% confidence interval 0.83 to 1.06).

    During childhood, increasing fatness was related to increased risk of coronary heart disease, particularly in those who were initially thin.1 This corresponds to our interaction model.2 The hazard ratio for the change in body mass index from age 1 to age 12 is obtainable from our combined model. The hazard ratios for body mass index at ages 1 and 12 are 0.83 and 1.03 (table 4). On the assumption that they would be similar if fitted simultaneously, the square root of the ratio of hazard ratios gives the approximate hazard ratio for the change in body mass index, Graphic—not that different from 1.20.

    So we agree that infant thinness and subsequent increasing fatness are synergistic risk factors for coronary heart disease, as others have shown.3 But for centile crossing to relate to coronary heart disease the hazard ratios for body size at the start and end of the period should differ significantly, and this is not the case in infancy. Routine use of our approach2 would have avoided this confusion.

    References

    1. 1.
    2. 2.
    3. 3.

    Authors' reply

    1. C Osmond, medical statistician (co{at}mrc.soton.ac.uk),
    2. D J P Barker, professor of clinical epidemiology,
    3. J G Eriksson, senior researcher,
    4. T Forsén, research fellow
    1. Centre for Paediatric Epidemiology and Biostatistics, Institute of Child Health, London WC1N 1EH
    2. MRC Childhood Nutrition Research Centre, Institute of Child Health
    3. MRC Environmental Epidemiology Unit (University of Southampton), Southampton General Hospital, Southampton SO16 6YD
    4. National Public Health Institute, Department of Epidemiology and Health Promotion, Diabetes and Genetic Epidemiology Unit, FIN-00300 Helsinki, Finland

      EDITOR—Cole et al are wrong. Coronary heart disease is clearly related to low weight gain during infancy in addition to low birth weight. Conditional on birth weight, the additional predictive power of infant weight gain is expressed by a χ2 statistic of 9.26 (P=0.002). In a simultaneous analysis the hazard ratio for a one standard deviation decrease in birth weight is 1.29 (95% confidence interval 1.14 to 1.45, P<0.001) and for a one unit decrease in standard deviation scores for weight between birth and age 1 it is 1.21 (1.08 to 1.36, P=0.001). The mistake that Cole et al make is in parameterising the model so that part of the effect of infant weight gain is lost in an average weight term.

      It is not adequate to analyse data on birth weight and weight at age 1 using what they describe as key regression models. These are dependent on assumptions of linearity. In the analyses of data from Hertfordshire, which first established the link between coronary heart disease and low weight gain in infancy, it was necessary to develop a more complicated model and express the results by using contours of disease risk.1 The Helsinki study provides a striking replication of these results and also allows us to examine the effects of growth through childhood. In our paper we focused on the finding that the effects of childhood weight gain on later coronary heart disease are conditioned by ponderal index at birth (birth weight/length3).

      Because the Helsinki dataset includes an average of nine measurements of height and weight during infancy for 8760 men and women we can now pinpoint the time in infancy when growth faltering begins and relate this to infant feeding, housing conditions, family size, and other variables. The study allows, for the first time, detailed description of the paths of fetal, infant, and childhood growth that precede the development of chronic diseases in later life. When these descriptions are published, would-be commentators on our analyses will be welcome to have any additional data needed for clarification. This will avoid the kind of erroneous conclusions that have been drawn by Cole et al.

      References

      1. 1.