Physiology and its Importance for Reference Intervals

Abstract

Reference intervals are ideally defined on apparently healthy individuals and should be distinguished from clinical decision limits that are derived from known diseased patients. Knowledge of physiological changes is a prerequisite for understanding and developing reference intervals. Reference intervals may differ for various subpopulations because of differences in their physiology, most obviously between men and women, but also in childhood, pregnancy and the elderly. Changes in laboratory measurements may be due to various physiological factors starting at birth including weaning, the active toddler, immunological learning, puberty, pregnancy, menopause and ageing. The need to partition reference intervals is required when there are significant physiological changes that need to be recognised. It is important that laboratorians are aware of these changes otherwise reference intervals that attempt to cover a widened inter-individual variability may lose their usefulness. It is virtually impossible for any laboratory to directly develop reference intervals for each of the physiological changes that are currently known, however indirect techniques can be used to develop or validate reference intervals in some difficult situations such as those for children. Physiology describes our life’s journey, and it is only when we are familiar with that journey that we can appreciate a pathological departure.

Introduction

Clinical investigation classically begins with gathering symptoms or examining for signs. The purpose is to identify the clinical features that aren’t typically found in healthy people as well as to detect those characteristics associated with a particular illness. Similarly, measurements made by clinical laboratories only have value when they can be compared to the values that are outside the usual spread of values found in health or within the spread of values typically found in disease. It is important these are two distinct questions i.e. ‘Is there evidence that the patient is not healthy?’ or ‘Is there evidence that a patient has a particular disease?’. Consequently we have separated two classes of thresholds that can be used by the clinical laboratory.^1–5

The first class of thresholds are reference intervals which describe the typical range of results seen in a healthy reference population. These were historically known as the ‘normal ranges’ but this term has been formally identified as incorrect and superseded according to the international standard for quality in medical laboratories (ISO 15189).⁶ The second class of thresholds are called clinical decision limits, where values above or below this threshold are considered diagnostic for the presence of a specific disease or are associated with a significant higher risk of the adverse clinical outcome(s). The most obvious example is a fasting glucose ≥7.0 mmol/L or a HbA₁c ≥6.5% (48 mmol/mol) being defined as decision limits for the diagnosis of diabetes mellitus based on associated clinical outcomes, including diabetic retinopathy,^7–9 although this decision point has been debated.^10–12

Reference intervals are defined with a high specificity for health (typically 95% or more) while clinical decision limits also consider sensitivity for disease. Receiver Operator Characteristic (ROC) curves are now widely used to balance the need for sensitivity and specificity. An ‘optimal’ cut-off derived using this technique may be neither a highly specific (95%) reference intervals nor a clinical decision limit based on high sensitivity. Whether ROC derived optimal limits, by balancing sensitivity and specificity, truly represent the best option can be debated according to clinical circumstances which may place higher importance on sensitivity or specificity. Therefore ‘optimal’ limits derived from ROC curves should be considered as an intermediate category of threshold which is neither a highly specific reference interval nor a sensitivity focussed clinical decision limit.

This article will focus on reference intervals, which are more widely used than clinical decision limits or ROC based cutoffs. It will also discuss the physiological aspects that are pre-requisite for defining reference intervals or agreeing on harmonised reference intervals. These physiological aspects are at least as important as the statistical aspects of reference intervals that are best obtained from the CLSI C28-A3 standard.¹³ This standard represents over 30 years of professional development in the theory and application of reference values largely developed by the International Federation of Clinical Chemistry (IFCC)^14–20 which was integral in developing that CLSI standard.

The considerable resources and statistical effort that are required to define reference intervals may be wasted if they don’t consider the underlying clinical purpose of reference intervals. In the current age of personalised medicine, the aim is to understand each patient as an individual. For every investigation our question is ‘What do I expect as a result for this particular patient if he/she is healthy?’

In terms of biological variability theory, when group inter-individual variability (CV_g) is much larger than the intra-individual variability (CV_i), reference intervals are less useful for judging individual patients.^21–25 Ideally the reference interval shouldn’t be much wider than each patient’s expected variations and if the ratio of CV_i to CV_g (the index of individuality) is below 0.6, reference intervals lose their utility.²⁶ More recently Petersen et al. have showed that the influence of the index of individuality on usefulness of reference intervals is even more important when a second sample is taken to confirm an abnormality.²⁷ Utility can be restored by stratifying (or partitioning) patients into similar groups. As exemplified by Fraser,²⁸ urine creatinine has an index of individuality of 0.46 when viewed as a whole and a urine creatinine reference interval is not as sensitive, for example, when a man’s urine creatinine level falls to the lower values normally seen in women. Separating urine creatinine reference intervals into gender based limits reduces inter-individual variation and improves the index of individuality to above 1.4 (1.42 for women and 1.83 for men), a value that confirms the utility of reference intervals.²⁶

As well as providing some background on the important physiological impacts on laboratory measurement from the literature, this paper will demonstrate that such physiological changes are also evident in the data generated by clinical laboratories and laboratory databases. While laboratory data is affected by results from diseased individuals, indirect techniques for investigating reference intervals can be used according to CLSI C28-A3,¹³ which states ‘the (indirect) techniques are perhaps more appropriately employed using data from individuals who are relatively healthy’. Furthermore, it is important to note that even known disease generally does not affect all analytes.²⁹ Over the last several years, the Australian laboratories of the Sonic Healthcare pathology network have been involved in a project to harmonise the reference intervals across these Australian laboratories.^30–34 The databases consist of a predominantly primary care population which is largely Caucasian, using common analytical techniques (Roche Modular biochemistry and Sysmex haematology). In our deliberations we have found that the changes in the population medians for these investigations reflect the major physiological changes already described in the literature as well as many subtle physiological changes that should also be considered when establishing reference intervals.

Physiology of Gender Based Reference Intervals

There are differences between men and women that cannot be disputed but, in terms of physiology, what are the factors behind those differences? The chromosomal differences between women and men are relatively small (46XX vs 46XY), yet they lead to profound sexual differences including the gonads, genitalia, breasts, hair and muscle. Each of these differences is largely understood at a biochemical level, from the impact of anti-mullerian hormone on the development of genitourinary tract in men,^35,36 to the effect of sex steroids on pubertal development.^37–41 There are other significant changes in biochemical tests that appear at puberty and are therefore attributable to these hormonal changes (Table 1). Haemoglobin and serum urate show a similar rise at puberty, but only in boys (Figure 1).

Table 1.

Changes in some common analytes at puberty and the sex hormones most likely to have caused the change.

Analyte	Pubertal Change	Sex Steroid
Creatine kinase	Boys rise by 50 IU/L	Testosterone
Creatinine	Boys rise by 15 μmol/L	Testosterone
Albumin	Boys rise by 2 g/L	Testosterone
Haemoglobin	Boys rise by 20 g/L	Testosterone
Urate	Boys rise by 0.05 mmol/L	Testosterone
Cholesterol	Boys fall by 0.4 mmol/L	Testosterone
Globulin	Girls rise by 2 g/L	Oestradiol
Platelets	Girls rise by 25 × 106/L	Oestradiol
Bicarbonate	Girls fall by 1.5 mmol/L	Progesterone

Figure 1.

The increase in haemoglobin (squares, full line) and urate (circle, dotted lines) in girls (grey) and boys (black) between the ages of 10 and 18 years. For haemoglobin there were 45,939 girls and 33,361 boys and for urate there were 30,164 girls and 23,444 boys.

Physiology of Childhood

Children are not little adults and reference intervals for any paediatric biomarker should be developed specifically for children and include well-recognised developmental changes.⁴² While sexual characteristic changes across puberty are profound, the earlier changes during the growth of a child, from birth to puberty are also significant. One of the most important tools in confirming a healthy child is the growth chart.⁴³ The growth represented in a child’s height or weight is not linear and typically has two growth spurts; one in the toddler age group (1 to 3 years) and one at adolescence (age 9 to 13 for girls and 11 to 16 for boys). Serum alkaline phosphatase (ALP) also demonstrates these two peaks, with the age of onset for the adolescent growth peak being earlier in pubertal girls and than pubertal boys (Figure 2).^44,45 Other bone markers also showing these differences.⁴⁶ This understanding is very important in setting paediatric reference intervals for alkaline phosphatase, as elevations of this enzyme can be considered the most common biochemical abnormality in Ricket’s⁴⁷ and important⁴⁸ or essential for its diagnosis.⁴⁹

Figure 2.

The changes in alkaline phosphatase plotted as the median value for each age group in 42,725 girls (grey) and 38,402 boys (black) derived from a laboratory population of predominantly outpatient children having a multiple biochemical analysis (screening) protocol.

These periods of bone growth also correspond to changes in calcium and phosphate metabolism (Figure 3). The higher calcium and phosphate levels during the ‘toddler’ growth surge has no gender differences, whereas the fall in phosphate levels during adolescence is earlier in girls than boys, mimicking the gender related delay in skeletal growth.⁵⁰

Figure 3.

The changes in serum calcium (cresolpthalien complexone) (squares, full line) and serum phosphate (circles, dotted lines) plotted as the median value for each age group in 42,725 girls (grey) and 38,402 boys (black) derived from a laboratory multiple biochemical analysis population of predominantly outpatient children.

Skeletal height is the most obvious measure of childhood growth, but this is also accompanied by changes in muscle mass. Serum creatinine is a marker of muscle mass in children (especially as their renal function is usually intact), and it is interesting to note the changes in serum creatinine in children (Figure 4); the trends show a gender related increase in creatinine that parallels the adolescent rise in haemoglobin. However, we can see that both creatinine and haemoglobin show an almost linear increase from 6 months to 12 years, without any gender differences in that period.^51–56 It is important to note that before puberty there aren’t any childhood surges in the rise of creatinine or haemoglobin that may be related to skeletal growth peaks, and therefore the most likely drivers of these changes are increasing physical activity and increased oxygen delivery requirements.

Figure 4.

The increase in median creatinine (Roche rate blanked modified Jaffe) (squares, full line) and median haemoglobin (circle, dotted lines) in girls (grey) and boys (black) between the ages of 10 and 18 years. For haemoglobin there were 62,971 girls and 48,289 boys and for urate there were 42,725 girls and 38,391 boys.

It is well known that respiratory rate and heart rate fall during childhood,^57,58 therefore how can we reconcile this with increasing physical activity and oxygen demand? One of the answers comes from looking at the bicarbonate changes in childhood (Figure 5). The rising serum bicarbonate in childhood has been described in many studies^59–61 but perhaps not fully appreciated as the consequence of a gradual fall in respiratory rate across childhood^57,58 and a continuing increase in absolute oxygen requirement. This leads to rising pCO₂ across childhood^62,63 which will result in rising bicarbonate.^64–66 Furthermore, due to the issues of electrolyte charge, the increase in bicarbonate (by 5 to 7 mmol/L) has influence on the more subtle changes in other electrolytes across in childhood such as sodium which rises by 2 to 3 mmol/L (Figure 5) and chloride (not shown in Figure 5) which falls by 2 to 3 mmol/L.⁶⁷ As these measured ion differences are balanced, they are not the cause of changes in the rising anion gap in childhood (rises in infancy and falls from age 2). This rising anion gap is caused by the changes in ‘unmeasured’ ions, especially rising albumin in infancy and falling phosphate.

Figure 5.

The increase in serum bicarbonate (squares, solid lines) and serum sodium (circles, dotted lines) in 42,725 girls (grey) and 38,391 boys (black) during childhood. The rise in childhood is related to the fall in respiratory rate across childhood and consequent increase in pCO₂, the source of serum bicarbonate.

Some metabolic changes in childhood seem to occur at the time of weaning. While digestive changes such as the reduction in lactase and increase in sucrase may be programmed for the expected change in diet,⁶⁸ metabolic programming may also be affected by nutritional experiences such as formula feeding.⁶⁹ The gut microbiota has also been implicated in the relationship between diet and metabolism.⁷⁰ The serum calcium pattern in infants and toddlers⁶⁰ shown in Figure 3 reflects the understood nutritional importance of milk. It is also interesting to look at the changes in cholesterol and triglycerides (Figure 6). While cholesterol rises at weaning (typically 6 months) and stays high, triglycerides rise at the same time but then fall back by age 5 years.⁶⁰

Figure 6.

The increase in median values for random serum cholesterol (squares, solid lines) triglycerides (circles, dotted lines) in 42,725 girls (grey) and 38,391 boys (black) during childhood. While both cholesterol and triglyceride rise in infancy, cholesterol stays high while triglycerides fall back by age 3.

Finally, the exposure to different antigens in food as well as different microbes represents one of the most important ‘behind the scenes’ changes in childhood; the training of the immune system. During weaning and in the toddler age group, neutrophils and globulins rise to their plateau at age 5,⁷¹ whereas lymphocytes have an enormous peak in infancy and then gradually fall to adult levels (Figure 7).⁷² The rise in globulins can be shown to be due to the rise in immunoglobulins.⁷³

Figure 7.

The changes in serum globulin (biuret protein – BCG albumin) (squares, solid line), neutrophil count (circles, dotted lines) and lymphocyte counts (stars, dashed line) in girls (grey) and boys (black) during childhood. For serum globulin there were 42,275 girls and 38,391 boys and for neutrophil and lymphocyte count there were 62,971 girls and 48,289 boys.

Physiology of Pregnancy

There are many important hormonal changes in pregnancy that ultimately impact on numerous aspects of physiology, such as the expansion in extracellular fluid volume⁷⁴ and corresponding increase in renal filtration.⁷⁵ These fluid changes are largely responsible for the typical falls in the concentration of most serum constituents. There are very few clinical laboratory measurements that rise in normal pregnancy, the most obvious being those related to the rise in oestrogens and progesterone. While we know that the synthetic oestrogens in the oral contraceptive pill increase transferrin levels and that the rise in oestrogens in pregnancy may do the same, most of the transferrin rise in pregnancy is in the third trimester and it is therefore misleading to attribute this solely to oestrogen. Furthermore, it is also know that iron stores are often depleted at this stage of rapid growth.⁷⁶

The rise in alkaline phosphatase in pregnancy is largely due production of placental alkaline phosphatase.⁷⁷ While most laboratorians are aware that alkaline phosphatase is higher in pregnant women, during the first and second trimester alkaline phosphatase levels are actually lower than in non-pregnant women and it is mainly in the third trimester that placental growth results in higher alkaline phosphatase levels.^78,79 This increase is mimicked by another placental product, Cystatin C,⁸⁰ which similarly undermines its usual clinical diagnostic role for renal function. An important biochemical measurement in the third trimester of pregnancy is serum urate, as increases are an important risk marker of pre-eclampsia.⁸¹ Obstetricians are aware that urate normally rises in the third trimester and risk thresholds change depending on the stage of the third trimester (Figure 8).⁸²

Figure 8.

The changes in serum urate (squares, solid line) and alkaline phosphatase (circles, dotted line) in 30,321 pregnant women of varying gestational age.

Figure 5 clearly shows that the median serum bicarbonate level in young women is 26 mmol/L and approximately 2 mmol/L lower than young men (28 mmol/L). This difference is because young women generally have lower pCO₂ than men.⁸³ In pregnancy, the pCO₂ and bicarbonate levels are even lower,⁸⁴ with bicarbonate a further 2 mmol/L lower than in non-pregnant women. All these changes are known to be due to the effect of progesterone which increases respiratory activity in pregnancy.⁸⁵

Physiological Changes in Adults

The most profound physiological transition in adulthood is the menopause. Hormonal changes in the menstrual cycle affect breathing⁸⁶ and, not surprisingly, the loss of the respiratory stimulation by oestrogen and progesterone ‘allows’ postmenopausal women to respire at a similar rate to men. Postmenopausal women adapt to the higher levels of pCO₂^87,88 unless hormone (especially progesterone) replacement therapy is given.⁸⁹ Serum bicarbonate levels correspondingly rise by 1 or 2 mmol/L in postmenopausal women. Whilst such subtle changes in bicarbonate are seldom of any clinical concern, the physiological importance of this menopausal change could be clinically important, since a rise in bicarbonate results in an increase in complexed calcium⁹⁰ and will increase the filtered renal load of calcium⁹¹ which, combined with a decrease in renal calcium reabsorption at menopause,⁹² can result in increased renal calcium loss at the menopause. Inevitably this will have to be replaced by diet or resorbed from bone.^93,94 The changes in median bicarbonate, calcium and alkaline phosphatase in serum shown in Figure 9 seem to be directly related to menopause and these physiological changes deserve much closer attention when we are trying to understand the physiology of the menopause and reference intervals across the climacteric.

Figure 9.

The changes in median serum bicarbonate (squares, solid line), median serum calcium (cresolpthalein complexone) (circles, dashed line) and median serum alkaline phosphatase (stars, dotted lines) in 74,032 women from 40 to 60 years of age.

None of these menopausal changes has a corresponding change in men. However men do have a gradual loss of the gender related differences that appeared at puberty. The gradual age-related decline in haemoglobin is much more obvious in men and can be related to the decline in testosterone levels with age.^95,96

Some may argue that any decline in function with human ageing is due to the accumulation of pathology rather than a physiological phenomenon.⁹⁷ As much as investigators have tried to understand ageing with concepts such as ‘inflammaging’,⁹⁸ telomere shortening,⁹⁹ oxidative damage,¹⁰⁰ or hormonal deficiency and metabolic decline,¹⁰¹ the explanation of ageing remains elusive. It is possible that human design has a built in ‘expiry date’¹⁰² (or at least a ‘best before date’). Studies in the healthiest elderly show significant changes including an age related decline in respiratory function (e.g. falling pO₂),^103–105 renal function (falling eGFR)^106–108 and cardiac status (rising high sensitivity troponin levels).^109–113 Many of the non-hormonal changes that occur with ageing are subtle compared to the hormonal changes such as for dehydroepiandrosterone sulphate (DHEAS).¹¹⁴ Whether these deteriorations represent normal ageing or the accumulation of pathology may be ‘academic’, because either way they may represent the known increasing health risk in the elderly. Age related reference intervals have the effect of ‘normalising’ physiological decline through maintaining age related specificity (i.e. 95%), but this necessarily also results in a decrease in sensitivity for disease and the associated mortality risks of ageing.

The Challenge of Partitioning

The partitioning of reference intervals into separate subclasses according to age, gender, ethnicity or ‘other’ is advisable when a clinical foundation or a logical physiological basis exists.¹¹⁵ Partitioning is a valuable tool for enhancing the diagnostic power of reference intervals.¹¹⁶ The partitioning of reference intervals is important but may also be the most complicated part of defining reference intervals. The partitioning into male and female seems the easiest step, but when should this be done? If the answer is where there is a significant difference between genders, what do we mean by significant difference? Differences have been generally considered as statistical differences. If there was a statistically significant difference for serum sodium of 1 mmol/L, how confident can we be of the usefulness of that difference in any particular individual when our routine assays cannot distinguish that difference? The balance between analytical quality and the quality of clinical interpretation is described in the Stockholm hierarchy,¹¹⁷ where the ultimate measure of quality is established by its relationship with clinical outcome. This ideal goal of defining analytical quality goals based on clinical outcome has not yet been applied to most measurements in laboratory medicine, let alone to the related issue of the impact of the quality of reference intervals on clinical interpretation and outcome.¹¹⁸ Therefore, as we are also unable to determine whether differences in the quality of partitioning of reference intervals will impact on clinical outcomes, we need to apply lesser approaches to the suitability of partitioning such as clinical opinion, statistics or laboratory consensus. Statistical partitioning methods by Stinton et al,¹¹⁹ Harris and Boyd¹²⁰ and Ichihara and Kawai,¹²¹ are essentially based on an arbitrary distance between two distributions, although improvements can be made by applying a prevalence adjusted distance.¹¹⁶ These methods don’t apply where there are more than two partitions to compare (as with almost all measurands). Gender can be assessed as two partitions, but what of age or pregnancy or where there is a continuous change? In order to compare the appropriate partitions, knowledge of the physiological changes affecting that measure is required.

As previously mentioned, when intra-individual variability is much tighter than inter-individual variability (index of individuality is below 0.6), reference intervals lose their usefulness. However, by partitioning reference intervals (to reduce inter-individual variation), the ratio of intra- to inter-individual variation can be increased above 1.4.

The dominant form of partitioning applied in clinical laboratory medicine is by social consensus: adulthood begins at age 18 (or 21), gestational age in pregnancy is divided into three trimesters, adult age can be divided into decades and old age is the age of retirement which is about 65 to 70 years. In this review, I hope to have demonstrated in the preceding discussions that while physiological changes often coincide with social and commonly used partitions such as weaning, adolescence, pregnancy, menopause and retirement, many of these physiological changes do not follow such partitions. For example, does physiology of pregnancy have any relevance to our arbitrary division of the nine months of pregnancy into three trimesters? Partitioning in adolescence should ideally be linked to the pubertal Tanner stages, but this requires considerable effort for laboratorians to develop this understanding as well as an effort by clinicians to apply these partitions. Childhood definitions are particularly problematic as, for example, thyroid stimulating hormone (TSH) falls to adult levels by age 12, while ALP falls to adult levels by age 21 in boys. It is difficult to create a set of partitions that reflects all the changes in childhood and this is probably why many reference interval studies in children settle on dividing childhood in 5 year blocks. The social divisions of childhood: newborn, infant, toddler, preschool, primary school and secondary school, are also more understandable for anxious parents who may otherwise be confused when trying to understand nuances of their child’s pathology reports. While there may be pragmatic reasons to apply these social divisions, as well as tools to decide whether these partitions lead to statistically significant differences, we should consider in these discussions, the scientific understanding of how physiology affects reference distributions.

Conclusion

Reference intervals should represent our understanding of physiology and the way it normally affects laboratory tests. This understanding is vital for maintaining the high specificity of reference intervals. The lack of a full appreciation of the importance of this understanding of physiology became obvious to our pathology network over the many years we have spent reviewing our reference intervals using direct and indirect reference interval data. We were fortunate to have access to hundreds of thousands of patient results that are largely an outpatient population where screening using multiple biochemical analysis is common and satisfies the CLSI C28-A3 requirement that indirect techniques are most appropriate using data from individuals who are relatively healthy.¹³ Indeed it would be nearly impossible to create corresponding data using expensive direct reference interval projects. If we need 120 healthy people to develop one direct reference interval, that requirement becomes 240 if there are gender differences, 360 more for the three trimesters of pregnancy and hundreds more if we are to consider age related partitioning for the elderly. We would also require hundreds more for the various stages of childhood¹²² which is why CLSI C28-A3 specifically states that indirect techniques ‘are used when it is deemed too difficult to collect samples from healthy subjects (e.g. paediatrics)’¹³ and investigators have used the indirect approach successfully in both paediatrics¹²³ and the elderly.¹²⁴ It is, therefore, virtually impossible to perform direct reference interval studies with enough individuals to represent all the physiological differences that are known to exist.¹²⁵ Indirect reference intervals usually compare very well with those derived directly,¹²⁶ but may reveal previously unsuspected differences.¹²⁷

As stated by Fraser: ‘appreciation of the biological changes that occur over the span of life is a necessary prerequisite to deciding whether stratification of reference values according to age is likely to be necessary’.¹²⁸ Human physiology describes our expected journey through life. We cannot define pathological departures from that journey without first understanding the journey itself.