Special issue: Quality in laboratory diagnostics: from theory to practice

Transcription

1 Special issue: Quality in laboratory diagnostics: from theory to practice Reference intervals as a tool for total quality management Gian Cesare Guidi, Gian Luca Salvagno* Clinical Chemistry Section, Department of Life and Reproduction Sciences, University of Verona, Italy *Corresponding author: gsalvagno77@yahoo.it, gianluca.salvagno@univr.it Abstract The more traditional, widespread and practiced method for interpreting the laboratory results is based on the comparison made with reference intervals. Nevertheless, the creation of appropriate reference intervals requires careful planning, monitoring and documentation of every aspect of the study, including the selection of the reference population (encompassing selection of homogeneous groups of reference according to ethnicity, geographical origin and environmental conditions, stratification according to age and gender, definition of health status) along with the use of the most appropriate statistical tools. In the very next future, the longitudinal comparison of laboratory results might probably replace the current use of reference intervals. Key words: reference intervals; reference range; statistics; laboratory diagnostics Re cei ved: Mar ch 23, 2010 Ac cep ted: May 1, 2010 Introduction The more traditional, widespread and practiced method for interpreting the laboratory results is based on the comparison made with reference intervals. As defined by the International Federation of Clinical Chemistry (IFCC) (1), the terms reference ran ge or inter val of re fe ren ce (IR) mean a range of values obtained from individuals (usually, but not necessarily healthy) randomly chosen, but appropriately selected in order to satisfy suitably defined criteria (2). The apparent contradiction between random and appropriate selection is resol ved bea ri ng in mi nd that the se are two pha ses of the sa me pro ce ss of iden ti fi ca tion of IR (3). Population based studies are preliminary steps for selecting the reference intervals, being numerous the va riab les that can a fe ct the po pu la tion characteristics (Table 1). It appears however very impor ta nt to stre ss the fa ct that ea ch la bo ra to ry should be able to establish the reference values that are as clo se as pos sib le to tho se pre sen ted by the population who insists close to the operating lab it se lf. Of cour se, it is un der stan dab le that in the eve nt of the in tro duc tion of new ac ti vi ties (new laboratory or new tests by any laboratory), as well as during transitional periods, provisional IR may be chosen, e.g. by opportunely adapting IRs from laboratories operating in nearby areas, as well as from reliable data in literature (4). However, the constraint should be of finding and setting own IRs as soon as pos sib le. In this cir cum stan ce, we Tab le 1. Preconditions for the formulation of a reference interval in healthy subjects. All re fe ren ce grou ps of in di vi dua ls shou ld be de fi ned. The patients studied should be similar to reference individuals for all aspects except those under consideration. Pre- and analytical conditions should be known. The quantities compared should be homogeneous. All results are derived from standardized methods in a system of adequa te ana lyti cal qua li ty. The stage of the disease process should be established. The diagnostic sensitivity and specificity, prevalence and clinical risk of misclassification should be known in advance. 165

2 shou ld thi nk about the in fluen ce that the prog ressi ve agi ng of the po pu la tion or the di fe re nt gender distribution observed in di f erent areas of the coun try ha ve on so me ana lysis (e.g., blood glu cose, blood urea nit ro gen, crea ti ni ne), or even mo re so the e fec ts of im mig ra tion from coun tries ve ry far or di f erent each other. IR establishment The creation of reference intervals requires careful planning, monitoring and documentation of every aspect of the study. Consequently, the reference intervals must be well characterized in terms of variations attributable to the pre-analytical and analytical factors (5). These formal protocols are particularly useful in cases where a laboratory should establish its own reference range for a particular test. This situation can occur even if a laboratory has mo di fied a te st or a met hod ap pro ved and/or certified, or a method developed in-house. Unfortu na te ly, the se pro to co ls are re sour ce in ten si ve and can be prohibitive for smaller facilities, also in consideration of the inherent costs (6). Even large laboratories may find it di f icult to carry out these stu dies for ob taining their own IRs, main ly ba sed on considerations of cost-benefit analysis. Thus many laboratories have increased their reliance on manufacturers to adopt reference intervals that may be acceptable using simpler approaches, which requi re le ss e fo rt and re su lt in lower cos ts. In any case, it is desirable that each laboratory has complete knowledge of the characteristics of the reference ranges adopted, such that they ensure compatibility with its own population and are suitable for clinical use. An IR is usually determined by analyzing samples that are ob tai ned from in di vi dua ls who meet the criteria previously and accurately defined (reference sample group). Protocols such as those made by the International Federation of Clinical Chemistry [IFCC] Expert Panel on Theory of Reference Va lues and by the Na tio nal Com mit tee for Cli nical Laboratory Standards show in a comprehensive and syste ma tic man ner the pro ces ses that use carefully selected reference sample groups to establish reference intervals. These protocols typically require a minimum of 120 reference persons for ea ch group (or sub group) to be cha rac te ri zed (7). However, it is increasingly gaining importance a vision that considers more appropriate to adopt reference intervals common to several laboratories that ope ra te over lar ge re gio nal areas and al so on entire national context (8 10). When establishing reference intervals that are common to mo st la bo ra to ries in the sa me area, the sample size can be expanded considerably around the local production of reference intervals for each individual laboratory. When many laboratories can share common reference intervals, the investment is limited and the whole operation can advantageous ly be con cen tra ted in one or a few in sti tu tions. Con sequen tly, one can wo rk on mu ch lar ger sam ple si zes, su ch as five/six hun dreds or mo re individuals. A larger sample makes it possible to carry out a thorough investigation of possible subgrou ps (11) in whi ch it is pos sib le to ob tain re liab le estimates on the reference intervals subgroup, respecting the minimum size of 120 individuals recommended by the IFCC. The confidence interval (CI) of 90% for a sam ple of si mi lar si ze is Cl = ± 0.24 SD (standard deviation of the population). The allocation criteria are (4): If one or bo th; the di fe ren ce be tween the lower reference limits and the di f erence between the hig her re fe ren ce li mi ts of the two subgrou ps are > 0.75 SD min (whe re SD min is the smal le st DS of the sub set of the DS), then the partition is recommended. If bo th; the di fe ren ces be tween the lower re ference limits and the higher reference limits of bo th sub grou ps are 0.25 SD min, then the partition is not recommended. For di fe ren ces whi ch fa ll be tween the extremes (0.25 SD min < di fe ren ce < 0.75 SD min ), the arguments should di f er from the purely sta tis ti cal ones, as this cou ld be due to ge ne tic di fe ren ces, i.e. to si tua tio ns whi ch are not routinely assessed. Selection of the reference population The selection of the population who will represent the re fe ren ce can not be dea lt wi th in ge ne ral 166

3 ter ms, as mo re than one va riab le ha ve to be con side red. The mo st com mon met hod is to ob tain reference values from a population of healthy individua ls, but in this ca se the de fi ni tion of heal th is indeed problematic. For example, to establish a reference interval for hemoglobin levels (i.e., a gender-related laboratory test), the laboratory would need to ob tain the re sul ts of he mog lo bin from at lea st 240 per so ns (120 men and 120 wo men). These peop le are usual ly drawn from the lo cal po pu lation and then se lec ted for in clu sion in the stu dy using carefully defined criteria. The general criteria that are adop ted are tho se re por ted in Table 2; moreover there is the opportunity to use a series of strategies, assuming additional criteria of subdivi sion for sub grou ps (Tab le 3) and/or age (Tab le 4) or combine multiple criteria, as for example (4): Selection of homogeneous groups of reference according to ethnicity, geographical origin and environmental conditions in order to obtain the representation of the population to which the normal range will apply. Stratification according to age and gender, if there are women pregnant or taking any anticonceptional drug. Definition of health status, according to further criteria that are adopted. There are no particular recommendations on which met hod of se lec tion is the mo st ap prop ria te, as this may de pe nd bo th on the pur po se of the inves ti ga tion, and on the op por tu ni ties al lowi ng to in clu de sin gle in di vi dua ls. In any ca se it is im porta nt to re po rt the stra te gy adop ted and the in divi dua ls in clu ded in the re fe ren ce in ter val and to implement clear criteria for inclusion and exclusion. Statistics The nor mal or Gaus sian dis tri bu tion (Figure 1) is the distribution characterized by two parameters, mean and standard deviation (SD). The statistical methods that assume Gaussian distribution of data are called parametric methods. Of course, other probability distributions, whose characteristics are de fi ned by one or mo re pa ra me te rs, can be analyzed using appropriate parametric methods. Tab le 2. Exclusion criteria for the formulation of a reference ran ge in the ge ne ral po pu la tion. Ri sk fac to rs Obe si ty Hypertension Risk factors related to environment and workplace Genetic risk factors Spe ci fic physio lo gi cal sta tes Preg nan cy Stress Exercise Dru gs Ge ne ric dru gs, oral con tra cep ti ves, al co hol, to bac co, etc. Tab le 3. Criteria for the creation of subgroups of reference subjec ts. Age (not necessarily with equal intervals of width) Gen der Ge ne tic Fac to rs Et hni ci ty Blood group (AB0) Histocompatibility antigens (HLA) Ge nes Spe ci fic physio lo gi cal sta tes ova rian cycle (hor mo nes) ges ta tio nal age physi cal con di tion Other fac to rs (so cioe co no mic, en vi ron men tal, chro no bio lo gi cal fac to rs) Tab le 4. Re fe ren ce in ter va ls: Cri te ria for dis tri bu tio ns in the diffe re nt age grou ps. Neo na tal pe riod (1 6 mon ths) Infan cy (6 mon ths 3 yea rs) Chil dhood (3 6 yea rs) Pre-pu ber tal (6 11 yea rs) Pu ber ty (11 18 yea rs) Adul thood (18 45 yea rs) Pre-me no pau sal Po st-me no pau sal Ma tu ri ty (45 65 yea rs) Old age (> 65 yea rs) 167

4 f(x) SD Mean SD Fi gu re 1. The normal or Gaussian distribution. x Non-parametric statistical techniques are used to ana lyze the da ta not ha vi ng a spe ci fic type of probability distribution. In general, when observing no n-gaus sian dis tri bu tio ns (no n-nor mal) (Figure 2a-b), their description is assigned to other indices such as median, percentiles classes, and more others. Moreover, in this second category of data distribution, other methods become more useful, inclu di ng the so cal led and im por ta nt ones bootstrap methods. Sometimes non-gaussian distributions can be normalized via appropriate processi ng tec hniques (12). This is the ge ne ral ca se of distributions obtained from experimental data, for which the assumption of normality is always verified. In constructing a reference range from individual da ta, of ten the di f icul ty of ac hie vi ng a perfect Gaussian distribution is apparent. Even after sam pling the da ta from a po pu la tion whi ch is presu med to be nor mal ly dis tri bu ted, it is of ten necessary to take some approximations of the data to com ply wi th the as sum ption of nor ma li ty. In this re ga rd a se ries of sta tis ti cal tes ts ha ve been put in place, which compares the distribution of experimental data with a hypothetical Gaussian distribution (13 15). The se met ho ds are cal led mat he matical-statistical goodness-of-fit test tests. Among them, the mo st known and used is the Kol mo gorov-smirnov, although its real discriminant power is questioned by some researchers, especially when the parameters of the distribution are estimated based on data rather than being specified a priori. Afterwards, other tests have been proposed that are be st sui ted for this pur po se, amo ng them the test of Shapiro-Wilks (for distribution of samples grea ter than 2,000 sub jec ts it shou ld be rep laced by the te st for nor ma li ty of Step hen) and the test of D Agostino-Pearson. None of these tests can however indicate the type of non-normality ob ser ved in the ca se whe re the dis tri bu tion is showing tendency to asymmetry (skewness) and kur to sis or bo th (Figure 3). The skewne ss rep re sents the deg ree of asymmet ry of a dis tri bu tion around its mean and is no n-di men sio nal sin ce it is characterized only by a number describing the shape of the distribution curve. When a distribution is per fec tly Gaus sian, the skewne ss sco re is a) b) f(x) f(x) Mode Median Mean Mean Median Mode x x Fi gu re 2. Non-Gaussian distribution (non-normal). 168

5 (+) Positively Skewed Distribution ( ) Negatively Skewed Distribution Fi gu re 3. Tendency to asymmetry (skewness). equal to 0. Skewness figures more or less negative or po si ti ve (e.g., +2.0 or 1.5) cor res po nd to a fo rm of dis tri bu tion cur ve wi th a tail mo re or le ss pronounced towards positive or negative values on the x axis. Similarly, the kurtosis represents the deg ree to whi ch the peak of dis tri bu tion is sha rp or flat, fluc tua ti ng be tween +3 and 3. In a per fectly Gaus sian cur ve the kur to sis sco re is 0 and of the dis tri bu tion is cal led me sokurtic (Figure 4). Many mathematical functions to correct either the skewne ss or the kur to sis have been pro po sed, and in some cases recommended, but their application was generally marginal. In practice, since a certain deg ree of skewne ss is always ob ser ved, a ru le of thumb has been defined according to which each distribution is considered Gaussian when the relationship between skewness and standard error is < ± 2. A si mi lar exer ci se is sug ges ted for the kur tosis, using the relationship between kurtosis and standard error of kurtosis. After ascertaining that the as sum ption of nor ma li ty is not vio la ted in a significant manner, the main parameters of the Gaussian curve (mean and standard deviation) are calculated and the interval of reference is considered to be com pri sed wit hin the va lues of the mean ± 1.96 stan da rd de via tion (so me ti mes 1.96 is rounded to 2.00) (Figure 5). When the as sum ption of nor ma li ty tes ts do not fit a normal curve, a logarithmic transformation of da ta can be used, in or der to res to re the da ta to a normal distribution curve; the above parameters (mean and SD) can be then cal cu la ted. Howe ver, so me ti mes no tran sfor ma tion and/or pro ces si ng of da ta is pos sib le. This can hap pen with da ta from mea su res of ana lytes expres sed by specific genes, such as highly polymorphic proteins (eg haptoglobin, lipoprotein (a)), homocysteine and others. To overcome these problems, the IFCC through its Expert Panel on Theory of Reference Values, has recommended the use of interpercentile intervals estimated on statistical methods either parametric or nonparametric, although the recommendation is in favour of the non-parametric (+) Leptokurtic ( ) Platykurtic (0) Mesokurtic (normal) 3S 2S S 68% S 2S 3S 95% 99,7% Fi gu re 4. General forms of kurtosis. Fi gu re 5. In ter val of re fe ren ce: Gaus sian cur ve. 169

6 approach (7). Although parametric methods are most commonly employed and seemingly simple from the poi nt of view of cal cu la tio ns, they maintain unresolved all the problems outlined above. The nonparametric methods, though a bit less easy to set up, ha ve the abi li ty to lar ge ly avoid su ch problems. Many procedures have been described (16). Cur ren tly the pre fer red met hod is ba sed on iterative bootstrap ranking (17). The target range is be tween the 2.5 th and the 97.5 th percentile (Figure 6). Even in the se ca ses the va lues be low and abo ve these limits are considered out of normality. A widely di f used but not supported by solid bases opi nion is that the re fe ren ce in ter val from Gaussian and no n-gaus sian dis tri bu tio ns rep re se nts the va lues of in di vi dua ls to be re fer red to (i.e., the nor mal in di vi dua ls ) and that the areas at the tails of the curve represent individuals whose values are to be re jec ted as out of nor ma lity. This is a misconception, because (18): 1. Even these values come from individuals originally included in the group chosen according to the characteristics set out before the construction of the interval of reference. 2. All va lues, bo th cen tral and tho se clo se to the limits of distribution, are only representations of biological variability on time. 3. In any case the analytical variability influences the current data. The abo ve con cep ts are we ll known to pro fes sionals in laboratories, but are largely ignored or underestimated in the clinical practice. Actually, the reference limits are not cut-of limits, because they 2.5 th percentile Median 97.5 th percentile Fi gu re 6. In ter per cen ti le in ter va ls: non pa ra met ric dis tri bu tion. are influenced by both the biological variability and the analytical one. Based on these considerations, the IFCC recommends estimating a confiden ce ran ge of 90% for ea ch limit of the re fe ren ce interval in both Gaussian and non-gaussian distribution. Longitudinal comparison of laboratory results The con ce pt of chan ge of re fe ren ce (CR) was proposed by Harris and Yasaka to enable evaluation of the observed change between two successive measures (19). The longitudinal comparison is based on this con ce pt and is main ly jus ti fied by the clinical problems that are not adequately answered by a cro ss-com pa ri son ba sed on the in ter val of reference. The Reference Change Value (or RCV) is especially useful in monitoring and follow-up of various clinical conditions. RCV is calculated by taking into account intra-individual biological variability, in addition to analytical variability in the medium to lo ng te rm, in or der to ta ke in to ac cou nt the ti me elap sed be tween the te st re sul ts. The general formula is as follows: RCV = z p 2 (CV a + CV w ); whe re z p is the pro ba bi li ty den si ty fun ction (gene ral ly 1.96 at P = 0.05), CV w is the in tra-in di vi dual bio lo gi cal va ria bi li ty and CV a is the variability of ana lyti cal tes ti ng. RCV shows so me spe cial ad ditio nal be ne fi ts to get in for ma tion on the sta tus of pa tien ts, par ti cu lar ly in the mo ni to ring of cli ni cally stab le and we ll con trol led con di tio ns, su ch as the prog no sis of the cri sis of re jec tion in kid ney tran spla nt pa tien ts, mo ni to ri ng of oral an ti coagu la nt the ra py (OAT), the glyca ted he mog lo bin (A1c) in dia be tes and ot her con di tio ns (20 23). RCV is on ly ap pli cab le when CV a < 0.5 < CV w. Close mo ni to ri ng of ana lyti cal qua li ty is needed for, es pe cial ly when the ti me be tween the fir st te st and the next is rat her lo ng su ch as for glyca ted he mog lo bin. In dis cus si ng the com pa ri son of lon gi tu di nal da ta it ap pea rs ap prop ria te to in tro du ce the con ce pt of in dex of in di vi dua li ty (I.I.) (24,25). The I.I. rep resen ts the ra tio of the ran dom dis tri bu tion of va 170

7 lues ob ser ved in sam ples ta ken from one in di vidual for a gi ven te st com pa red to the dis tri bu tion of va lues of the en ti re po pu la tion of in di vi dua ls for the sa me te st. When the ob ser ved I. I. is low, it is of lit tle cli ni cal uti li ty usi ng tra di tio nal re fe rence in ter val. A cu t-of va lue of I.I. 0.6 is con si dered and in this ca se the com pa ri son of lon gi tu dinal da ta is mu ch mo re sui ted to eva lua te the chan ges ob ser ved usi ng RCV. When the re sul ts of la bo ra to ry tes ts wi th a low RCV are lo ca ted near the li mi ts of dis tri bu tion of the tra di tio nal IR, in a position of low frequency, there are two possibilities: a) stab le con di tion if the pre vious re su lt was simi lar; b) a con di tion ac hie ved in re ce nt ti mes if the result show va riation. Since in this case the traditional IR is insensitive and therefore not needed, only a previous result of that te st can cla ri fy the si tua tion. It is al so important to consider that many of the laboratory tests that explore aspects of body metabolism show low ho meos ta tic I.I. in res pe ct of IR. For a number of tests it seems therefore important to collect the results in databases or systems for collecting personal data (such as chip-based flash memory card, now widely available and very inexpen si ve) to ac ce ss when nee ded. For eve ry repe ti tion of the se ries of tes ts, the re sul ts shou ld be col lec ted, and com pa red to the pre vious ones. Recently the concept of estimating the di f erences be tween se rial re sul ts as the pro ba bi li ty of change by calculating the likelihood ratio (likelihood ra tio) in ad di tion to RCV has been in tro duced (26). The pro ce du re ap pea rs ro bu st from a theo re ti cal poi nt of view and de ser ves to be wide ly adop ted, as it see ms li ke ly to im pro ve the mo ni to ri ng of in di vi dual con di tio ns and pro vi de cli ni cal sup po rt to ra tio nal cli ni cal de ci sion. Ea ch in di vi dual cou ld be ne fit from a prog res si ve as sessme nt ( in prog re ss ) of their own heal th and any deviation from her/his reference state identified and assessed. Conclusions The quality performances resulting from the current technology advancements allow clinical laboratories to fully exploit the opportunity of creating common IRs in order to accomplish transferability of data, thus increasing citizens benefits and meeting health system expectations. References 1. Solberg HE. The IFCC recommendation on estimation of refe ren ce in ter va ls. The Ref Val prog ram. Clin Chem Lab Med 2004;42: Solberg HE, PetitClerc C. International Federation of Clinical Chemistry (IFCC), Scientific Committee, Clinical Section, Expert Panel on Theory of Reference Values. Approved recommendation (1987) on the theory of reference values. Part 2. Selection of individuals for the production of referen ce va lues. J Clin Chem Clin Bioc hem 1987;25: Solberg HE, PetitClerc C. International Federation of Clinical Chemistry (IFCC), Scientific Committee, Clinical Section, Expert Panel on Theory of Reference Values. Approved recommendation (1988) on the theory of reference values. Part 3. Preparation of individuals and collection of specime ns for the pro duc tion of re fe ren ce va lues. J Clin Chem Clin Bioc hem 1988;26: Hyltoft Petersen P, Rustad P. Prerequisites for establishing common reference intervals. Scand J Clin Lab Invest 2004;64: Solberg HE, Stamm D. International Federation of Clinical Chemistry (IFCC), Scientific Committee, Clinical Section, Expert Panel on Theory of Reference Values. Approved recommendation on the theory of reference values. Part 4. Control of analytical variation in the production, transfer and ap pli ca tion of re fe ren ce va lues. Eur J Clin Chem Clin Biochem 1991;29: Fuentes-Arderiu X, Ferré-Masferrer M, González-Alba JM, Escolà-Aliberas J, Balsells-Rosello D, Blanco-Cristobal C, et al. Multicentric reference values for some quantities measured with Tina-Quant reagents systems and RD/Hitachi ana lyse rs. Sca nd J Clin Lab In ve st 2001;61: Sol be rg HE. The theo ry of re fe ren ce va lues Pa rt 5. Sta tisti cal treat me nt of col lec ted re fe ren ce va lues. De ter mi nation of re fe ren ce li mi ts. J Clin Chem Clin Bioc hem 1983;21: Ferré-Masferrer M, Fuentes-Arderiu X, Alvarez-Funes V, Güell-Miró R, Castiñeiras-Lacambra MJ. Multicentric reference va lues: sha red re fe ren ce li mi ts. Eur J Clin Chem Clin Biochem 1997;35: Rus tad P, Fel di ng P, Fran zson L, Kai ris to V, Lah ti A, Mår tensson A, et al. The Nor dic Re fe ren ce In ter val Pro je ct 2000: recommended reference intervals for 25 common biochemical pro per ties. Sca nd J Clin Lab In ve st 2004;64:

8 10. Rustad P, Felding P, Lahti A, Hyltoft Petersen P. Descriptive analytical data and consequences for calculation of common reference intervals in the Nordic Reference Interval Pro je ct Sca nd J Clin Lab In ve st 2004;64: Strømme JH, Rus tad P, Steen sla nd H, Theo dor sen L, Ur dal P. Reference intervals for eight enzymes in blood of adult females and males measured in accordance with the International Federation of Clinical Chemistry reference system at 37 deg rees C: pa rt of the Nor dic Re fe ren ce In ter val Pro ject. Sca nd J Clin Lab In ve st 2004;64: Hyltoft Petersen P, Blaabjerg O, Andersen M, Jøngersen LGM, Schousboe K, Jensen E. Graphical interpretation of confiden ce cur ves in ran kit plo ts. Clin Chem Lab Med 2004;42: D Agostino RB. An omnibus test of normality for moderate and large sample sizes. Biometrika 1971;58: Pearson E, D Agostino RB, Bowman K. Tests for departure from normality: comparison of powers. Biometrika 1977; 64: D Agostino RB, Stephens MA. Goodness-of-fit techniques. New Yo rk: Mar cel Dek ker, I nc; Reed AH, Hen ry JR, Ma son WB. In fluen ce of sta tis ti cal method used on the re sul ti ng es ti ma te of nor mal ran ge. Clin Chem 1971;17: Harris EK, Boyd JC. Statistical bases of reference values in laboratory medicine. New York: Marcel Dekker, Inc; Guidi GC, Lippi G, Solero GP, Poli G, Plebani M. Managing transferability of laboratory data. Clin Chim Acta 2006; 374: Harris EK, Yasaka T. On the calculation of a Reference Change for comparing two consecutive measurements. Clin Chem 1983;29: Biosca C, Ricòs C, Lauzurica R, Galimany R, Hyltoft Petersen P. Reference change value concept combining two delta values in predict crises in renal posttransplantation. Clin Chem 2001;47: Las sen JF, Kjel dsen J, An ton sen S, Hylto ft Pe ter sen P, Brandslund I. Interpretation of serial measurements of international normalized ratio for prothrombin times in monitoring oral anticoagulant therapy. Clin Chem 1995;41: Lassen JF, Brandslund I, Antonsen S. International normalized ratio for prothrombin times in patients taking oral anticoagulants: critical di f erence and probability of significant change in consecutive measurements. Clin Chem 1995;41: Skeie S, Pe ri ch C, Ri còs C, Arac zki A, Hor va th AR, Oos ter huis WP, et al. Postanalytical external quality assessment of blood glucose and hemoglobin A1c: an international survey. Clin Chem 2005;51: Harris EK. E f ects of intra- and interindividual variation on the ap prop ria te use of nor mal ran ges. Clin Chem 1974;20: Hylto ft Pe ter sen P, Fra ser CG, San dbe rg S, Goldschmi dt H. The index of individuality is often a misinterpreted quantity cha rac te ris tic. Clin Chem Lab Med 1999;37: Hyltoft Petersen P, Sandberg S, Iglesias N, Sölétormos G, Aasne KA, Brandslund I, Jørgensen GM. Likelyhood-ratio and odds applied to monitoring of patients as a suppleme nt to re fe ren ce chan ge va lue (RCV). Clin Chem Lab Med 2008;46; Referentni interval kao alat u upravljanju kvalitetom Sažetak Metoda za tumačenje laboratorijskih rezultata koja je tradicionalnija, rasprostranjenija i češće u uporabi temelji se na usporedbi referentnim intervalima. Međutim, određivanje odgovarajućeg referentnog intervala zahtjeva pažljivo planiranje, praćenje i dokumentiranje svakog aspekta istraživanja uključujući odabir referentne populacije (odabir homogene skupine prema etničkoj pripadnosti, zemljopisnom porijeklu i društvenom okruženju, stratifikacija prema dobi i spolu, definicija zdravstvenog stanja), zajedno s primjenom najprimjerenijeg statističkog alata. U vrlo bliskoj budućnosti longitudinalne usporedbe laboratorijskih rezultata mogle bi zamijeniti primjenu referentnih intervala na način na koji ih danas primjenjujemo. Ključ ne ri je či: referentni intervali; referentni raspon; statistički testovi; laboratorijska dijagnostika 172