On 2014 Nov 23, Stephen Wood commented:

We feel the authors should consider the possibility that their key finding, that Large for Gestational Age (LGA) birth weight is a risk factor for stillbirth, was an artifact of sampling. We note that 50% of the stillbirths in your study occurred at <28 weeks gestation yet < 1% of the live birth controls were <28 weeks and only 1% <32 weeks. This may have been appropriate for the analysis that determined LGA using population norms but error may have been introduced when using the ultrasound (Hadlock) or indvidualized norms. In developing these norms Hadlock et al assessed only 392 subjects and although they had ultrasounds at a range of gestational ages the accuracy of the estimates was only reported for deliveries > 36 weeks. Therefore, it is uncertain how accurate they would be for preterm fetuses. The individualized norms are also prone to similar error as they are as essentially slightly adjusted Hadlock ultrasound norms. As your controls were a population based sample not matched by gestational age then this error could have been important. To investigate this possibility we analyzed data from our provincial perinatal database. For the years 1992-2009 there were 727693 singleton births at 23 weeks or greater gestation, we excluded 11 with gestational age>45 weeks and 3216 with missing birth weights. In the first analysis all stillbirths were matched with the next two live births (population based controls). The gestational age distribution of live birth controls was similar to your study with only 1.1% delivering <28 weeks. Using the same population norms as your study (Alexander) we observed similar results to those you reported; SGA was associated with stillbirth using both population and ultrasound norms but LGA only increased the risk of stillbirth using the Hadlock ultrasound norms. In fact using the population norms it was protective.

Hadlock U/S norms SGA (small for gestational age) OR 6.38 (5.77, 7.06) AGA (appropriate for gestational age) Reference LGA (large for gestational age) OR 1.91 (1.64, 2.22)

Alexander Population norms SGA OR 6.43 (5.74, 7.21) AGA Reference LGA OR 0.61 (0.51, 0.74)

To evaluate the possible effect of sampling we then matched all stillbirths to the next two live births with the same gestational age.

Hadlock (ultrasound norms) SGA OR 4.77 (4.32, 5.26) AGA Reference LGA OR 0.98 (0.85, 1.13)

Alexander (population norms) SGA OR 6.47 (5.78, 7.24) AGA Reference LGA OR 0.83 (0.69, 1.01)

The results as you see are quite different. While SGA is still strongly associated with stillbirth LGA is not using either population or ultrasound norms. This suggests that the increase in risk of LGA observed in your study was an artifact of sampling. This almost certainly applies to the results using the "individualized norms" as they essentially modification of the ultrasound norms. We suggest that you do a similar analysis on your own data with gestational age matched controls to ensure the increase risk of stillbirth you observed with LGA is robust to changes in the control population.

On 2014 Nov 23, Stephen Wood commented:

We feel the authors should consider the possibility that their key finding, that Large for Gestational Age (LGA) birth weight is a risk factor for stillbirth, was an artifact of sampling. We note that 50% of the stillbirths in your study occurred at <28 weeks gestation yet < 1% of the live birth controls were <28 weeks and only 1% <32 weeks. This may have been appropriate for the analysis that determined LGA using population norms but error may have been introduced when using the ultrasound (Hadlock) or indvidualized norms. In developing these norms Hadlock et al assessed only 392 subjects and although they had ultrasounds at a range of gestational ages the accuracy of the estimates was only reported for deliveries > 36 weeks. Therefore, it is uncertain how accurate they would be for preterm fetuses. The individualized norms are also prone to similar error as they are as essentially slightly adjusted Hadlock ultrasound norms. As your controls were a population based sample not matched by gestational age then this error could have been important. To investigate this possibility we analyzed data from our provincial perinatal database. For the years 1992-2009 there were 727693 singleton births at 23 weeks or greater gestation, we excluded 11 with gestational age>45 weeks and 3216 with missing birth weights. In the first analysis all stillbirths were matched with the next two live births (population based controls). The gestational age distribution of live birth controls was similar to your study with only 1.1% delivering <28 weeks. Using the same population norms as your study (Alexander) we observed similar results to those you reported; SGA was associated with stillbirth using both population and ultrasound norms but LGA only increased the risk of stillbirth using the Hadlock ultrasound norms. In fact using the population norms it was protective.

Hadlock U/S norms SGA (small for gestational age) OR 6.38 (5.77, 7.06) AGA (appropriate for gestational age) Reference LGA (large for gestational age) OR 1.91 (1.64, 2.22)

Alexander Population norms SGA OR 6.43 (5.74, 7.21) AGA Reference LGA OR 0.61 (0.51, 0.74)

To evaluate the possible effect of sampling we then matched all stillbirths to the next two live births with the same gestational age.

Hadlock (ultrasound norms) SGA OR 4.77 (4.32, 5.26) AGA Reference LGA OR 0.98 (0.85, 1.13)

Alexander (population norms) SGA OR 6.47 (5.78, 7.24) AGA Reference LGA OR 0.83 (0.69, 1.01)

The results as you see are quite different. While SGA is still strongly associated with stillbirth LGA is not using either population or ultrasound norms. This suggests that the increase in risk of LGA observed in your study was an artifact of sampling. This almost certainly applies to the results using the "individualized norms" as they essentially modification of the ultrasound norms. We suggest that you do a similar analysis on your own data with gestational age matched controls to ensure the increase risk of stillbirth you observed with LGA is robust to changes in the control population.