Limitations of the Hoffmann method for establishing reference intervals using clinical laboratory data

To examine the effect of different limits of acceptability on the Hoffmann method for selecting "healthy" populations from laboratory test data.

Thyroid stimulating hormone (TSH) and free thyroxine (FT₄) were measured in Shunyi Maternal and Children's Hospital of Beijing Children's Hospital. The sample size of participants of TSH and FT₄ for reference intervals (RIs) establishment was 10,864 and 10,799, respectively. RIs were calculated by Hoffmann method with different acceptable deviations (α value). The validation data was collected prospectively and the out-of-range (OOR) values were calculated to examine the applicability of RIs with different acceptable deviations. The sample size for RIs validation was 880 and 867, respectively. The RIs were considered as valid when OOR was <10%.

α value was set at different levels for establishing the RIs of TSH and FT₄. It was shown that the larger α value, the wider the RI. The RIs calculated by the Hoffmann method, under the default α value of 0.05, were much narrower than the previous findings. The OOR of both TSH and FT₄ were far more from 10% when the α value was 0.05. In this simulation, the OOR of TSH and FT₄ was not <10% until the α value was set as 0.55 and 0.80, respectively. The established RIs were valid for both training dataset and validation dataset.

It can be somewhat subjective to define the acceptable deviation when selecting "normal" reference individuals. The default value of acceptable deviation may not be applicable in some cases. It is necessary to determine the acceptable deviation based on the certain condition, instead of using the default value directly.