Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 14-Day Trial for You or Your Team.

Learn More →

Demonstrating the requirements for new reference ranges for diagnosis and prognosis to reflect subsets of the population

Demonstrating the requirements for new reference ranges for diagnosis and prognosis to reflect... Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 BioscienceHorizons Volume 11 2018 10.1093/biohorizons/hzy016 ............................................................................................ ..................................................................... Research article Demonstrating the requirements for new reference ranges for diagnosis and prognosis to reflect subsets of the population Helen Wilson University of the West of England, Coldharbour Lane, Bristol BS16 1QY, UK *Corresponding author: University of the West of England, Coldharbour Lane, Bristol BS16 1QY, UK. Tel: +0117 328 3653. Email: Helen.Wilson@warwick.ac.uk Supervisor: Dr Antony Hill, University of the West of England, Coldharbour Lane, Bristol BS16 1QY, UK. ............................................................................................ ..................................................................... Reference ranges (RRs) are not produced for different subsets of the population, other than for gender. This study aimed to explore the impact of generalised RRs on diagnosis thus proving the necessity for new RRs taking into account different eth- nicities (hispanic, white, black and Asian). This impact was explored by direct comparisons of RRs produced for different eth- nicities in which significant differences would provide evidence for possible misdiagnosis. This was achieved by creating RRs from the National Health and Nutrition Examination Survey, a programme of studies assessing the health and nutritional sta- tus of children and adults within USA. The 2.5th and 97.5th percentiles from the 2013–2014 data was calculated after those deemed unhealthy were removed via a set of extensive parameters. Having bootstrapped the data 30 times, a Kruskal–Wallis statistical test was performed followed by a Mann–Whitney post hoc. Nearly all male lower and upper limits were found to be significantly different to each other, apart from the upper limits of hispanic to both white and black, and white to black. Female results showed almost an equal amount of significant and non-significant differences. Those which were significantly different were the upper and lower limits of all the ethnicities to Black, alongside the lower limits of White to Asian. The results of this study suggest that a proportion of misdiagnosis could be avoided if separate RRs were produced, at least for different ethnicities. Key words: haematocrit, ethnicity, reference ranges, population study, bootstrapping Submitted on 6 October 2018; editorial decision on 3 December 2018 ............................................................................................ ..................................................................... treatment may ensue. It is generally accepted that the lower Introduction th th and upper limits are represented by the 2.5 and 97.5 per- centiles respectively (Jones and Barker, 2008). Despite this, A total of 60–70% of all clinical decisions come from labora- the RRs in use do not account for differences between subsets tory results (Lim, Miyamura, and Chen, 2015). Therefore, it of the population i.e. ethnicity, apart from gender, as demon- is incredibly important that the reference ranges (RRs) strated by clinical RRs in Table 1 (Gloucestershire Hospitals, (defined as a range when applied to a population, and ser- 2015). viced by a laboratory correctly, includes most of the subjects with characteristics similar to the reference group whilst There is little research exploring the need for RRs to be set excluding the others (Ceriotti, 2007)) used by clinicians are as for different ethnicities, with only one study including the accurate as possible. Otherwise, misdiagnosis and incorrect Asian ethnicity in the analysis, (Lim, Miyamura, and Chen, ............................................................................................... .................................................................. © The Author(s) 2019. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Research article Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. Table 1. RRs from Gloucestershire Hospitals (2015) 2015). The guidelines by the International Federation of Clinical Chemistry advise that every country must establish Male haematocrit (%) Female haematocrit (%) their own RRs for health (Malati, 2009); however, these guidelines do not encourage the production of RRs for differ- 40–54 37–47 ent subsets of the population. In countries where one ethnicity Table 2. Parameters used for data selection for NHANES data 2013–2014 Demographics Cardiovascular disease Those within the ages 18–65 years old Those who are not taking medication for hypertension Data separated according to gender Those who are not taking medication for high blood pressure Current health status (self-reported) being good or above. Those who do not have high blood pressure Those who are not over weight Those who are not taking low dose aspirin Metabolic diseases Those who have not ever been told that they have angina Those who are not taking medication for high cholesterol. Those who have never been told that they have congestive heart failure Those who do not have high cholesterol Those who have not ever been told that they have coronary heart disease Those who have not been told by a doctor that they have diabetes Those who have not ever been told that they have had a heart attack Those who do not have high blood sugar Those who have not received a blood transfusion since 2000 Those who are not hypoglycaemic Those who have not ever been told that they have had a stroke Those who do not currently have a thyroid problem Those who are not taking treatment for anaemia Those who do not have a current liver condition Lifestyle choices Those who have not been told they have been jaundice Those who have not used marijuana in over 5 years Those who have not been told they have weak or failing kidneys Those who have not used cocaine in over 5 years Pulmonary diseases Those who have not used heroine in over 5 years Those who have not ever been told they have emphysema Those who have not used methamphetamine in over 5 years Those who do not currently have chronic bronchitis Those who have not inject drugs in over 5 years Those who have not ever been told they have chronic obstructive pulmonary Infectious disease disease Those who have not smoked for 2 years Those who have not been told they have hepatitis C Those who have not used tobacco or nicotine in the past 5 years Those who have not been told they have hepatitis B Those who do not have other smokers within the house (i.e. those who are not Cancerous diseases exposed to second hand smoking) Those who have not had cancer or any other malignancy ............................................................................................... .................................................................. 2 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. predominates, having a sole RR will affect the health of a experiment the participants’ medical history, along with a minority. However, in countries such as the UK and USA, self-report on health status, was used to remove those who where populations are highly diverse, limited RRs can have a would not reflect a healthy population. In particular, the far wider detrimental impact. Generalised RRs are a signifi- removal of any individuals with habits, such as smoking, cant problem in non-industrialised nations where clinicians which would affect circulation was ensured, along with those adopt Western World text books RRs which are heavily who were diseased. The parameters used for the 2013–2014 skewed towards a Caucasian population, as demonstrated NHANES data are shown in Table 2. Having gathered and th th with the scenario of 20% of Tanzanian children being diag- arranged the appropriate data the 2.5 and 97.5 percentiles nosed with adverse health effects if the US National Health were found to represent the lower and upper limits Division of Aids’ RRs were applied to their population (Lim, respectively. Miyamura, and Chen, 2015). It was felt that further investiga- Having calculated RRs for each gender of the four different tion was needed into Lim et al.’s study. The parameters used ethnicities (Hispanic, White, Black and Asian), statistical ana- by Lim et al. were not robust, relying on honest and accurate lysis could be carried out to establish whether there were any self-reporting of good health allowing some participants significant differences. The difficulty with statistical analysis within their sample who would not represent a healthy popu- of RRs is that only one value is produced for the lower and lation. Therefore, this study wanted to use more extensive upper limits. Having approached a statistician, Dr Ana parameters which used the participants’ medical records. Sendova-Franks of the University of the West of England, it Also, it was believed that a different statistical test may be was concluded that a suitable method to analyse the data more appropriate, as the quantile regression model demon- would be via bootstrapping it. Bootstrapping creates more strated whether there was a correlation between the ranges, data sets by random sampling of the original data with not whether there was a significant difference between them. replacement (Efron, 2003). This then creates 30 sets of data Therefore, it was decided a Mann–Whitney test would be for each of the male and female ethnicities, allowing a more appropriate carried out on bootstrapped data. The pro- Kruskal–Wallis statistical test followed by a Mann–Whitney ject’s aim was to confirm Lim et al.’s findings, specifically for post hoc test to be performed. Other studies have discussed haematocrit (the percentage of blood which is composed of the benefits of using bootstrapping where the data set is too RBCs (Hill and Jensen, 2019)), with newer data, more appro- small for meaningful statistics for the production of RRs priate data selection parameters and different statistical (Coskun et al. 2013; Ozarda, 2016). It has been demonstrated analysis. that 90% confidence intervals can be determined for small data sets using this method, however, for larger data sets, such as those used in this study, then 95% confidence inter- Materials and methods vals can be achieved. If significant differences were found between the RRs, it could be assumed that it may be possible for people to be incorrectly Results deemed unhealthy or vice versa. Whilst the Lim et al. paper gave RRs gathered from NHANES, they were created from The results of this study suggest that separate RRs should be data from 2011–2012 using less suitable parameters, possibly provided for different ethnicities according to gender. The resulting in skewed data; after all, RRs are supposed to be RRs produced from the 2013–2014 NHANES data is dis- representative of a healthy population (Lim, Miyamura, and played in Table 3, including the groups’ sample size. Nearly Chen, 2015). This study agreed with Lim et al. on the age par- all the male lower and upper limits were significantly different ameter used, 18–65 years old, as the young and old would to each other, apart from the upper limits of Hispanic to both skew laboratory test results. Similarly, this study approved of Black and White, and White to Black, as demonstrated in separating the data according to gender. Within this Fig. 1. The female results showed almost an equal amount of Table 3. RRs produced from NHANES 2013–2014 data after data selection parameters were applied, calculated as the 2.5th and 97.5th percentiles after separated by gender and ethnicity Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit (%) Hispanic 143 32.2–48.7 Hispanic 160 32.3–48.6 White 184 32.1–48.6 White 197 32.3–48.2 Black 88 34.2–48.3 Black 101 32.8–47.5 Asian 60 31.6–46.2 Asian 82 31.9–48.5 ............................................................................................... .................................................................. 3 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Research article Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. significant and non-significant differences. Those which were Unfortunately, the data that Lim et al. had gathered from significant were the upper limits of all the ethnicities to Black, NHANES (2011–2012) was not available, therefore, statis- alongside the lower limits of White to Asian, as demonstrated tical tests could not establish whether there were significant in Fig. 2. differences between the RRs produced by the two studies (Lim, Miyamura, and Chen, 2015). Nevertheless, just by sight, one would not be surprised the two studies’ RRs were all significantly different. Lim et al.’s data can be viewed in Table 4. However, this study did have access to the same data that Lim et al. did, i.e. the NHANES 2011–2012 data. Therefore, further investigations were performed to establish where differences lay between the RRs of Lim et al. and this study by producing two sets of RRs from the same data they used, one with their parameters and one with this study’s. Very interestingly, by using the rigorous parameters of this study, the majority of Lim et al.’s participants were excluded, for example, using the stricter parameters there were only 165 White male participants, compared to 565 used by Lim et al. Even more interesting, replicates of their RRs were unable to be produced despite using their parameters. The biggest dif- ference being the male Hispanic lower limits with a difference of 7.5%. This is even more intriguing as the sample size dif- Figure 1. Male reference limits (produced from bootstrapped NHANES data from 2013 to 2014) are significantly different to each fers in every group between the replicate and Lim et al.’s other according to a Mann–Whitney post hoc test. study, despite using the same parameters. Surprisingly, the biggest difference in participant number was 92 for Hispanic females. The differences in the study sample sizes can be seen in Table 4, showing Lim et al.’s data, and Table 5, showing the replication of their study. It could be considered that the population’s haematocrit values significantly changed between the dates of data acquisi- tion i.e. 2011–2012 to 2013–2014. This is why our study was replicated with the 2011–2012 data using our parameters. These data are shown in Table 6, where it can be seen that the RRs are similar. There is a small variation which could pos- sibly be due to a number of confounding factors, such as environmental change. Discussion Figure 2. Female reference ranges (produced from bootstrapped NHANES data from 2013 to 2014) are significantly different to each It was assumed that the RRs of Lim et al. would be considerably other according to a Mann–Whitney post hoc test. different to the ones that this study produced due to different Table 4. RRs produced from NHANES 2011–2012 data, which was separated for gender and ethnicity and the sample included those who were between the ages of 18 and 65 years with those who self-reported that their health was good or above Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit range (%) Hispanic 316 38.8–49.5 Hispanic 276 31.0–44.1 White 608 38.7–50.0 White 552 33.6–44.9 Black 425 36.1–49.6 Black 419 29.5–43.6 Asian 254 36.7–49.4 Asian 227 32.2–43.8 Data from Lim, Miyamura and Chen (2015) ............................................................................................... .................................................................. 4 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. Table 5 Replication of the study of Lim, Miyamura, and Chen (2015) to produce RRs from NHANES 2011–2012 data, which was separated for gender and ethnicity and the sample included those who were between the ages of 18–65 years with those who self-reported that their health was good or above Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit range (%) Hispanic 343 31.3–48.2 Hispanic 368 31.5–48.1 White 565 31.6–47.8 White 511 31.9–47.7 Black 428 31.6–48.2 Black 481 32.4–47.5 Asian 252 32.2–48.0 Asian 247 32.3–48.4 Table 6. RRs produced from NHANES 2011–2012 data after selection criteria were applied from Table 2 and then calculated from the 2.5th and 97.5th percentiles after the data were separated by gender and ethnicity Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit range (%) Hispanic 111 31.3–48.3 Hispanic 123 31.8–48.5 White 165 31.8–47.8 White 161 32.4–47.8 Black 118 31.9–47.7 Black 151 32.5–47.8 Asian 94 33.5–47.3 Asian 85 32.6–48.9 parameter choices (Lim, Miyamura, and Chen, 2015). Lim et al. the 2013–2014 RRs. The comparison in these results nowhere relied on honest participants for self-diagnosis of current health near reflects the differences seen between Lim’s data and this status to decide whether to include participants as representatives study, so this rules out differences between the 2011–2012 of a healthy population to make their RRs. However, health is and 2013–2014 participants as the cause. highly subjective and those who were smoking, or had other long As this study’s parameters were stricter, and data sets smal- term conditions, may have felt themselves healthy despite having ler, an outlier within the data set could have a bigger effect on different laboratory test results, such as haematocrit, to that of a skewing our data. As bootstrapping had already been carried population who are deemed healthy by virtue of their medical his- out for statistical testing it made sense to see whether sample tory. Therefore, it is highly likely that Lim et al.’s RRs were heav- size could have skewed the project’s results. To achieve this, ily skewed by these participants. This was demonstrated in the all the bootstrapped data for an ethnicity of a gender were comparison of Lim et al.’s parameters to those of this study using combined and new RRs produced, giving a sample size 30 the 2011–2012 NHANES data, where a majority of their partici- times bigger. Little difference was found, suggesting that out- pants were excluded when more rigorous parameters were liers did not skew the original data. In some groups, the same applied. This suggests that the self-reporting was inaccurate and limit was found, for example, the male white lower and upper that the majority of their data had come from people who did not limits, the male Asian lower limit, and the female white and represent a healthy population so should not have been included black upper limits. It made sense to use the RRs from the in the production of RRs. The difference in sample size between bootstrapped data for blood flow modelling as they would be the replication of Lim et al.’s study and the original must be due more accurate. The bootstrapped data is shown in Table 7. to human error. It is peculiar that Lim et al.’s study produced a lower limit of 36.7% for male Asians where the replication found Due to availability this study uses American data. The RRs 32.2%, especially when considering that there was only a differ- for this data therefore are only applicable to the USA and may ence of two participants between the groups. not reflect other populations. Differing factors such as air pol- lution, health care, a more severely obesogenic society or dif- As previously stated, differences between RRs could be ferences in regulation and advice on food, exercise and other caused by differences between the 2011–2012 data (which factors may mean that their representative healthy population Lim et al. used) and the 2013–2014 data (which this project may not match others. For the RRs produced within this used). Hence the recreation of our project and its parameters study to be considered truly valid, further subsets should be with the 2011–2012 data which could then be compared to ............................................................................................... .................................................................. 5 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Research article Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. Table 7. RRs produced from the bootstrapped 2013–2014 NHANES data after selection criteria was applied from Table 2 and then calculated from the 2.5th and 97.5th percentiles after the data was separated by gender and ethnicity Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit range (%) Hispanic 4290 32.4–48.5 Hispanic 4800 32.4–48.3 White 5520 32.1–48.6 White 5911 32.6–48.2 Black 2640 34.4–48.2 Black 3030 33.2–47.5 Asian 1800 31.6–46.0 Asian 2460 32.0–48.3 reflected. For example, haematocrit values are affected by age without his help. I would also like to thank Dr Ana Sendova- (Ujiie et al., 2009) and it is possible that within the NHANES Franks, who thankfully put me onto the right path with the survey a certain ethnicity had a higher average age than statistics. My project would have been completely incorrect if another, thus possibly skewing the RR. Further research not for her help. I would also like to thank Dr Eunjung Lim would be required to ensure that differences we found from the University of Hawaii at Manoa who showed me resulted solely from ethnicity. With the small sample size, a where I could access the raw data for my research. I would result from our parameters, it would not have been feasible to have never been able to find it if not for her help. Also, a huge further split our data by age as there would not be enough thanks for giving me permission to present her data for dis- data to be statistically relevant. cussion within this project. I would also like to thank the friends and family who helped throughout my project, espe- cially with proof reading, in particular Liz Wilson, my Conclusion mother, and Charlotte Dimond. The findings of this study suggest that there is a requirement for different RRs to be set, at least for diagnostic purposes. The RRs were produced from a large sample size suggesting Funding that the RRs produced in Table 7 would be valid for the None received. population the data was gathered from, i.e. the USA. Considering the benefits of producing RRs for different sub- sets of the population, new ones should be created to reflect the current population. This would reduce stress on the health References services with a reduction in incorrect diagnoses and unneces- Ceriotti, F. (2007) Prerequisites for use of common reference intervals, sary treatments. Also, from a patient perspective the sooner The Clinical Biochemist Reviews, 28, 115–121. correct diagnosis is made the sooner treatment can be given. This could dramatically alter their outcome, as with many Coskun, A., Ceyhan, E., Inal, T. C. et al. (2013) The comparison of para- conditions rapid diagnosis is vitally important. metric and nonparametric bootstrap methods for reference interval computation in small sample size groups, Accreditation and Quality Assurance, 18, 51–60. Author biography Efron, B. (2003) Second thoughts on the bootstrap, Statistical Science, Having studied Biological Sciences BSc (Hons) at the University 18, 135–140. of the West of England and discovering a love for Gloucestershire Hospitals (2015) Haematology Reference Ranges Microbiology, whilst missing Mathematics, Helen Wilson (accessed 4 February 2017). decided to specialise within the field of Mathematical Modelling through a MRes at the University of Birmingham with the hopes Hill, A. A. and Jensen, O. E. (2019) Modelling red blood cell transport in that this will continue into a PhD with her dream subject being the microcirculation, in preparation. the study of the probiotic nature of microbes. Jones, G. and Barker, A. (2008) Reference intervals, Clinical Biochemist Reviews, 29, 93–97. Acknowledgements Lim, E., Miyamura, J. and Chen, J. J. (2015) Racial/ethnic-specific refer- ence intervals for common laboratory tests: a comparison among First, I would like to thank my research supervisor, Dr Asians, blacks, hispanics, and white, Hawai’i Journal of Medicine & Antony Hill, whose guidance throughout this project has Public Health, 74, 302–310. been invaluable. I would not have ever got through the maths ............................................................................................... .................................................................. 6 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. Malati, T. (2009) Whether western normative laboratory values used for Ozarda, Y. (2016) Reference intervals: current status, recent develop- clinical diagnosis are applicable to the Indian population? An over- ments and future considerations, Biochemia Medica, 26 (1), 5–16. view on reference intervals, Indian Journal of Clinical Biochemistry, Ujiie, H., Kawasaki, M., Suzuki, Y. et al. (2009) Influence of age and hemato- 24, 111–122. crit on the coagulation of blood, Journal of Bioreheology, 23, 111–114. ............................................................................................... .................................................................. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png BioScience Horizons Oxford University Press

Demonstrating the requirements for new reference ranges for diagnosis and prognosis to reflect subsets of the population

BioScience Horizons , Volume 11 – Jan 1, 2018

Loading next page...
 
/lp/oxford-university-press/demonstrating-the-requirements-for-new-reference-ranges-for-diagnosis-3q7bmNJc39
Publisher
Oxford University Press
Copyright
Copyright © 2022 Oxford University Press
eISSN
1754-7431
DOI
10.1093/biohorizons/hzy016
Publisher site
See Article on Publisher Site

Abstract

Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 BioscienceHorizons Volume 11 2018 10.1093/biohorizons/hzy016 ............................................................................................ ..................................................................... Research article Demonstrating the requirements for new reference ranges for diagnosis and prognosis to reflect subsets of the population Helen Wilson University of the West of England, Coldharbour Lane, Bristol BS16 1QY, UK *Corresponding author: University of the West of England, Coldharbour Lane, Bristol BS16 1QY, UK. Tel: +0117 328 3653. Email: Helen.Wilson@warwick.ac.uk Supervisor: Dr Antony Hill, University of the West of England, Coldharbour Lane, Bristol BS16 1QY, UK. ............................................................................................ ..................................................................... Reference ranges (RRs) are not produced for different subsets of the population, other than for gender. This study aimed to explore the impact of generalised RRs on diagnosis thus proving the necessity for new RRs taking into account different eth- nicities (hispanic, white, black and Asian). This impact was explored by direct comparisons of RRs produced for different eth- nicities in which significant differences would provide evidence for possible misdiagnosis. This was achieved by creating RRs from the National Health and Nutrition Examination Survey, a programme of studies assessing the health and nutritional sta- tus of children and adults within USA. The 2.5th and 97.5th percentiles from the 2013–2014 data was calculated after those deemed unhealthy were removed via a set of extensive parameters. Having bootstrapped the data 30 times, a Kruskal–Wallis statistical test was performed followed by a Mann–Whitney post hoc. Nearly all male lower and upper limits were found to be significantly different to each other, apart from the upper limits of hispanic to both white and black, and white to black. Female results showed almost an equal amount of significant and non-significant differences. Those which were significantly different were the upper and lower limits of all the ethnicities to Black, alongside the lower limits of White to Asian. The results of this study suggest that a proportion of misdiagnosis could be avoided if separate RRs were produced, at least for different ethnicities. Key words: haematocrit, ethnicity, reference ranges, population study, bootstrapping Submitted on 6 October 2018; editorial decision on 3 December 2018 ............................................................................................ ..................................................................... treatment may ensue. It is generally accepted that the lower Introduction th th and upper limits are represented by the 2.5 and 97.5 per- centiles respectively (Jones and Barker, 2008). Despite this, A total of 60–70% of all clinical decisions come from labora- the RRs in use do not account for differences between subsets tory results (Lim, Miyamura, and Chen, 2015). Therefore, it of the population i.e. ethnicity, apart from gender, as demon- is incredibly important that the reference ranges (RRs) strated by clinical RRs in Table 1 (Gloucestershire Hospitals, (defined as a range when applied to a population, and ser- 2015). viced by a laboratory correctly, includes most of the subjects with characteristics similar to the reference group whilst There is little research exploring the need for RRs to be set excluding the others (Ceriotti, 2007)) used by clinicians are as for different ethnicities, with only one study including the accurate as possible. Otherwise, misdiagnosis and incorrect Asian ethnicity in the analysis, (Lim, Miyamura, and Chen, ............................................................................................... .................................................................. © The Author(s) 2019. Published by Oxford University Press. This is an Open Access article distributed under the terms of the Creative Commons Attribution License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted reuse, distribution, and reproduction in any medium, provided the original work is properly cited. Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Research article Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. Table 1. RRs from Gloucestershire Hospitals (2015) 2015). The guidelines by the International Federation of Clinical Chemistry advise that every country must establish Male haematocrit (%) Female haematocrit (%) their own RRs for health (Malati, 2009); however, these guidelines do not encourage the production of RRs for differ- 40–54 37–47 ent subsets of the population. In countries where one ethnicity Table 2. Parameters used for data selection for NHANES data 2013–2014 Demographics Cardiovascular disease Those within the ages 18–65 years old Those who are not taking medication for hypertension Data separated according to gender Those who are not taking medication for high blood pressure Current health status (self-reported) being good or above. Those who do not have high blood pressure Those who are not over weight Those who are not taking low dose aspirin Metabolic diseases Those who have not ever been told that they have angina Those who are not taking medication for high cholesterol. Those who have never been told that they have congestive heart failure Those who do not have high cholesterol Those who have not ever been told that they have coronary heart disease Those who have not been told by a doctor that they have diabetes Those who have not ever been told that they have had a heart attack Those who do not have high blood sugar Those who have not received a blood transfusion since 2000 Those who are not hypoglycaemic Those who have not ever been told that they have had a stroke Those who do not currently have a thyroid problem Those who are not taking treatment for anaemia Those who do not have a current liver condition Lifestyle choices Those who have not been told they have been jaundice Those who have not used marijuana in over 5 years Those who have not been told they have weak or failing kidneys Those who have not used cocaine in over 5 years Pulmonary diseases Those who have not used heroine in over 5 years Those who have not ever been told they have emphysema Those who have not used methamphetamine in over 5 years Those who do not currently have chronic bronchitis Those who have not inject drugs in over 5 years Those who have not ever been told they have chronic obstructive pulmonary Infectious disease disease Those who have not smoked for 2 years Those who have not been told they have hepatitis C Those who have not used tobacco or nicotine in the past 5 years Those who have not been told they have hepatitis B Those who do not have other smokers within the house (i.e. those who are not Cancerous diseases exposed to second hand smoking) Those who have not had cancer or any other malignancy ............................................................................................... .................................................................. 2 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. predominates, having a sole RR will affect the health of a experiment the participants’ medical history, along with a minority. However, in countries such as the UK and USA, self-report on health status, was used to remove those who where populations are highly diverse, limited RRs can have a would not reflect a healthy population. In particular, the far wider detrimental impact. Generalised RRs are a signifi- removal of any individuals with habits, such as smoking, cant problem in non-industrialised nations where clinicians which would affect circulation was ensured, along with those adopt Western World text books RRs which are heavily who were diseased. The parameters used for the 2013–2014 skewed towards a Caucasian population, as demonstrated NHANES data are shown in Table 2. Having gathered and th th with the scenario of 20% of Tanzanian children being diag- arranged the appropriate data the 2.5 and 97.5 percentiles nosed with adverse health effects if the US National Health were found to represent the lower and upper limits Division of Aids’ RRs were applied to their population (Lim, respectively. Miyamura, and Chen, 2015). It was felt that further investiga- Having calculated RRs for each gender of the four different tion was needed into Lim et al.’s study. The parameters used ethnicities (Hispanic, White, Black and Asian), statistical ana- by Lim et al. were not robust, relying on honest and accurate lysis could be carried out to establish whether there were any self-reporting of good health allowing some participants significant differences. The difficulty with statistical analysis within their sample who would not represent a healthy popu- of RRs is that only one value is produced for the lower and lation. Therefore, this study wanted to use more extensive upper limits. Having approached a statistician, Dr Ana parameters which used the participants’ medical records. Sendova-Franks of the University of the West of England, it Also, it was believed that a different statistical test may be was concluded that a suitable method to analyse the data more appropriate, as the quantile regression model demon- would be via bootstrapping it. Bootstrapping creates more strated whether there was a correlation between the ranges, data sets by random sampling of the original data with not whether there was a significant difference between them. replacement (Efron, 2003). This then creates 30 sets of data Therefore, it was decided a Mann–Whitney test would be for each of the male and female ethnicities, allowing a more appropriate carried out on bootstrapped data. The pro- Kruskal–Wallis statistical test followed by a Mann–Whitney ject’s aim was to confirm Lim et al.’s findings, specifically for post hoc test to be performed. Other studies have discussed haematocrit (the percentage of blood which is composed of the benefits of using bootstrapping where the data set is too RBCs (Hill and Jensen, 2019)), with newer data, more appro- small for meaningful statistics for the production of RRs priate data selection parameters and different statistical (Coskun et al. 2013; Ozarda, 2016). It has been demonstrated analysis. that 90% confidence intervals can be determined for small data sets using this method, however, for larger data sets, such as those used in this study, then 95% confidence inter- Materials and methods vals can be achieved. If significant differences were found between the RRs, it could be assumed that it may be possible for people to be incorrectly Results deemed unhealthy or vice versa. Whilst the Lim et al. paper gave RRs gathered from NHANES, they were created from The results of this study suggest that separate RRs should be data from 2011–2012 using less suitable parameters, possibly provided for different ethnicities according to gender. The resulting in skewed data; after all, RRs are supposed to be RRs produced from the 2013–2014 NHANES data is dis- representative of a healthy population (Lim, Miyamura, and played in Table 3, including the groups’ sample size. Nearly Chen, 2015). This study agreed with Lim et al. on the age par- all the male lower and upper limits were significantly different ameter used, 18–65 years old, as the young and old would to each other, apart from the upper limits of Hispanic to both skew laboratory test results. Similarly, this study approved of Black and White, and White to Black, as demonstrated in separating the data according to gender. Within this Fig. 1. The female results showed almost an equal amount of Table 3. RRs produced from NHANES 2013–2014 data after data selection parameters were applied, calculated as the 2.5th and 97.5th percentiles after separated by gender and ethnicity Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit (%) Hispanic 143 32.2–48.7 Hispanic 160 32.3–48.6 White 184 32.1–48.6 White 197 32.3–48.2 Black 88 34.2–48.3 Black 101 32.8–47.5 Asian 60 31.6–46.2 Asian 82 31.9–48.5 ............................................................................................... .................................................................. 3 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Research article Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. significant and non-significant differences. Those which were Unfortunately, the data that Lim et al. had gathered from significant were the upper limits of all the ethnicities to Black, NHANES (2011–2012) was not available, therefore, statis- alongside the lower limits of White to Asian, as demonstrated tical tests could not establish whether there were significant in Fig. 2. differences between the RRs produced by the two studies (Lim, Miyamura, and Chen, 2015). Nevertheless, just by sight, one would not be surprised the two studies’ RRs were all significantly different. Lim et al.’s data can be viewed in Table 4. However, this study did have access to the same data that Lim et al. did, i.e. the NHANES 2011–2012 data. Therefore, further investigations were performed to establish where differences lay between the RRs of Lim et al. and this study by producing two sets of RRs from the same data they used, one with their parameters and one with this study’s. Very interestingly, by using the rigorous parameters of this study, the majority of Lim et al.’s participants were excluded, for example, using the stricter parameters there were only 165 White male participants, compared to 565 used by Lim et al. Even more interesting, replicates of their RRs were unable to be produced despite using their parameters. The biggest dif- ference being the male Hispanic lower limits with a difference of 7.5%. This is even more intriguing as the sample size dif- Figure 1. Male reference limits (produced from bootstrapped NHANES data from 2013 to 2014) are significantly different to each fers in every group between the replicate and Lim et al.’s other according to a Mann–Whitney post hoc test. study, despite using the same parameters. Surprisingly, the biggest difference in participant number was 92 for Hispanic females. The differences in the study sample sizes can be seen in Table 4, showing Lim et al.’s data, and Table 5, showing the replication of their study. It could be considered that the population’s haematocrit values significantly changed between the dates of data acquisi- tion i.e. 2011–2012 to 2013–2014. This is why our study was replicated with the 2011–2012 data using our parameters. These data are shown in Table 6, where it can be seen that the RRs are similar. There is a small variation which could pos- sibly be due to a number of confounding factors, such as environmental change. Discussion Figure 2. Female reference ranges (produced from bootstrapped NHANES data from 2013 to 2014) are significantly different to each It was assumed that the RRs of Lim et al. would be considerably other according to a Mann–Whitney post hoc test. different to the ones that this study produced due to different Table 4. RRs produced from NHANES 2011–2012 data, which was separated for gender and ethnicity and the sample included those who were between the ages of 18 and 65 years with those who self-reported that their health was good or above Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit range (%) Hispanic 316 38.8–49.5 Hispanic 276 31.0–44.1 White 608 38.7–50.0 White 552 33.6–44.9 Black 425 36.1–49.6 Black 419 29.5–43.6 Asian 254 36.7–49.4 Asian 227 32.2–43.8 Data from Lim, Miyamura and Chen (2015) ............................................................................................... .................................................................. 4 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. Table 5 Replication of the study of Lim, Miyamura, and Chen (2015) to produce RRs from NHANES 2011–2012 data, which was separated for gender and ethnicity and the sample included those who were between the ages of 18–65 years with those who self-reported that their health was good or above Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit range (%) Hispanic 343 31.3–48.2 Hispanic 368 31.5–48.1 White 565 31.6–47.8 White 511 31.9–47.7 Black 428 31.6–48.2 Black 481 32.4–47.5 Asian 252 32.2–48.0 Asian 247 32.3–48.4 Table 6. RRs produced from NHANES 2011–2012 data after selection criteria were applied from Table 2 and then calculated from the 2.5th and 97.5th percentiles after the data were separated by gender and ethnicity Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit range (%) Hispanic 111 31.3–48.3 Hispanic 123 31.8–48.5 White 165 31.8–47.8 White 161 32.4–47.8 Black 118 31.9–47.7 Black 151 32.5–47.8 Asian 94 33.5–47.3 Asian 85 32.6–48.9 parameter choices (Lim, Miyamura, and Chen, 2015). Lim et al. the 2013–2014 RRs. The comparison in these results nowhere relied on honest participants for self-diagnosis of current health near reflects the differences seen between Lim’s data and this status to decide whether to include participants as representatives study, so this rules out differences between the 2011–2012 of a healthy population to make their RRs. However, health is and 2013–2014 participants as the cause. highly subjective and those who were smoking, or had other long As this study’s parameters were stricter, and data sets smal- term conditions, may have felt themselves healthy despite having ler, an outlier within the data set could have a bigger effect on different laboratory test results, such as haematocrit, to that of a skewing our data. As bootstrapping had already been carried population who are deemed healthy by virtue of their medical his- out for statistical testing it made sense to see whether sample tory. Therefore, it is highly likely that Lim et al.’s RRs were heav- size could have skewed the project’s results. To achieve this, ily skewed by these participants. This was demonstrated in the all the bootstrapped data for an ethnicity of a gender were comparison of Lim et al.’s parameters to those of this study using combined and new RRs produced, giving a sample size 30 the 2011–2012 NHANES data, where a majority of their partici- times bigger. Little difference was found, suggesting that out- pants were excluded when more rigorous parameters were liers did not skew the original data. In some groups, the same applied. This suggests that the self-reporting was inaccurate and limit was found, for example, the male white lower and upper that the majority of their data had come from people who did not limits, the male Asian lower limit, and the female white and represent a healthy population so should not have been included black upper limits. It made sense to use the RRs from the in the production of RRs. The difference in sample size between bootstrapped data for blood flow modelling as they would be the replication of Lim et al.’s study and the original must be due more accurate. The bootstrapped data is shown in Table 7. to human error. It is peculiar that Lim et al.’s study produced a lower limit of 36.7% for male Asians where the replication found Due to availability this study uses American data. The RRs 32.2%, especially when considering that there was only a differ- for this data therefore are only applicable to the USA and may ence of two participants between the groups. not reflect other populations. Differing factors such as air pol- lution, health care, a more severely obesogenic society or dif- As previously stated, differences between RRs could be ferences in regulation and advice on food, exercise and other caused by differences between the 2011–2012 data (which factors may mean that their representative healthy population Lim et al. used) and the 2013–2014 data (which this project may not match others. For the RRs produced within this used). Hence the recreation of our project and its parameters study to be considered truly valid, further subsets should be with the 2011–2012 data which could then be compared to ............................................................................................... .................................................................. 5 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Research article Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. Table 7. RRs produced from the bootstrapped 2013–2014 NHANES data after selection criteria was applied from Table 2 and then calculated from the 2.5th and 97.5th percentiles after the data was separated by gender and ethnicity Male Female Ethnicity Number in sample Haematocrit range (%) Ethnicity Number in sample Haematocrit range (%) Hispanic 4290 32.4–48.5 Hispanic 4800 32.4–48.3 White 5520 32.1–48.6 White 5911 32.6–48.2 Black 2640 34.4–48.2 Black 3030 33.2–47.5 Asian 1800 31.6–46.0 Asian 2460 32.0–48.3 reflected. For example, haematocrit values are affected by age without his help. I would also like to thank Dr Ana Sendova- (Ujiie et al., 2009) and it is possible that within the NHANES Franks, who thankfully put me onto the right path with the survey a certain ethnicity had a higher average age than statistics. My project would have been completely incorrect if another, thus possibly skewing the RR. Further research not for her help. I would also like to thank Dr Eunjung Lim would be required to ensure that differences we found from the University of Hawaii at Manoa who showed me resulted solely from ethnicity. With the small sample size, a where I could access the raw data for my research. I would result from our parameters, it would not have been feasible to have never been able to find it if not for her help. Also, a huge further split our data by age as there would not be enough thanks for giving me permission to present her data for dis- data to be statistically relevant. cussion within this project. I would also like to thank the friends and family who helped throughout my project, espe- cially with proof reading, in particular Liz Wilson, my Conclusion mother, and Charlotte Dimond. The findings of this study suggest that there is a requirement for different RRs to be set, at least for diagnostic purposes. The RRs were produced from a large sample size suggesting Funding that the RRs produced in Table 7 would be valid for the None received. population the data was gathered from, i.e. the USA. Considering the benefits of producing RRs for different sub- sets of the population, new ones should be created to reflect the current population. This would reduce stress on the health References services with a reduction in incorrect diagnoses and unneces- Ceriotti, F. (2007) Prerequisites for use of common reference intervals, sary treatments. Also, from a patient perspective the sooner The Clinical Biochemist Reviews, 28, 115–121. correct diagnosis is made the sooner treatment can be given. This could dramatically alter their outcome, as with many Coskun, A., Ceyhan, E., Inal, T. C. et al. (2013) The comparison of para- conditions rapid diagnosis is vitally important. metric and nonparametric bootstrap methods for reference interval computation in small sample size groups, Accreditation and Quality Assurance, 18, 51–60. Author biography Efron, B. (2003) Second thoughts on the bootstrap, Statistical Science, Having studied Biological Sciences BSc (Hons) at the University 18, 135–140. of the West of England and discovering a love for Gloucestershire Hospitals (2015) Haematology Reference Ranges Microbiology, whilst missing Mathematics, Helen Wilson (accessed 4 February 2017). decided to specialise within the field of Mathematical Modelling through a MRes at the University of Birmingham with the hopes Hill, A. A. and Jensen, O. E. (2019) Modelling red blood cell transport in that this will continue into a PhD with her dream subject being the microcirculation, in preparation. the study of the probiotic nature of microbes. Jones, G. and Barker, A. (2008) Reference intervals, Clinical Biochemist Reviews, 29, 93–97. Acknowledgements Lim, E., Miyamura, J. and Chen, J. J. (2015) Racial/ethnic-specific refer- ence intervals for common laboratory tests: a comparison among First, I would like to thank my research supervisor, Dr Asians, blacks, hispanics, and white, Hawai’i Journal of Medicine & Antony Hill, whose guidance throughout this project has Public Health, 74, 302–310. been invaluable. I would not have ever got through the maths ............................................................................................... .................................................................. 6 Downloaded from https://academic.oup.com/biohorizons/article/doi/10.1093/biohorizons/hzy016/5304539 by guest on 17 August 2022 Bioscience Horizons � Volume 11 2018 ............................................................................................... .................................................................. Malati, T. (2009) Whether western normative laboratory values used for Ozarda, Y. (2016) Reference intervals: current status, recent develop- clinical diagnosis are applicable to the Indian population? An over- ments and future considerations, Biochemia Medica, 26 (1), 5–16. view on reference intervals, Indian Journal of Clinical Biochemistry, Ujiie, H., Kawasaki, M., Suzuki, Y. et al. (2009) Influence of age and hemato- 24, 111–122. crit on the coagulation of blood, Journal of Bioreheology, 23, 111–114. ............................................................................................... ..................................................................

Journal

BioScience HorizonsOxford University Press

Published: Jan 1, 2018

References