Influence of Analytical Bias and Imprecision on Guideline-Driven Medical Decision Limits Per Hyltoft Petersen Hyltoft Petersen P, Klee GG. Influence of Analytical Bias and Imprecision on the Number of False Positive Results Using Guideline-Driven Medical Decision Limits. Clin Chem Acta 2014;430:1-8 LABMED 2014

“Guideline-Driven Medical Decision Limits”. Diagnostic decisions based on decision limits according to medical guidelines are different from the majority of clinical decisions due to the strict dichotomization of patients into diseased and non-diseased based on the biochemical measurement of a single component. Background : Consequently, the influence of analytical performance is more critical than for other diagnostic decisions where many other informations are included in the diagnosis.

Cholesterol as a screening-test for preventing coronary heart disease in adults Decision: Cholesterol above or below 6.2 mmol/L “Guideline-Driven Medical Decision Limits”. HbA1c in diagnosis of diabetes mellitus Decision: HbA1c above or below 48 mmol/mol (6.5 % HbA1c) Examples : Sacks et al. Diabetes Care 2011;34:c61-c99 Garber et al. Ann Intern Med ;124:

“Guideline-Driven Medical Decision Limits”. HbA1c reference interval for healthy According to traditional IFCC criteria CV WITHIN-SUBJECT = 1.94 % in IFCC units (~ ln = ) HbA1c : Recommended cut-off = 48 mmol/mol ~ ln = 3.86 Jørgensen et al. Scand J Clin Lab Invest 2002; 62: Carlsen et al. Clin Chem Lab Med 2011;49: Log-Gaussian distribution (natural logarithm) ln-mean = and ln-standard deviation = 0.053

“Guideline-Driven Medical Decision Limits”. Distribution of healthy set-points CV WITHIN-SUBJECT = 1.94 % HbA1c : Distribution of values from a person with set-point = cut-off

“Guideline-Driven Medical Decision Limits”. HbA1c : Distribution for a person with set-point = cut-off FrequencyCumulated frequency (probability) Probability of being measured above cut-off 50 %

“Guideline-Driven Medical Decision Limits”. HbA1c : Distribution for a person with set-point = cut-off Effect of one and two samplings

“Guideline-Driven Medical Decision Limits”. HbA1c : Distribution for two samplings with set-point = cut-off Effect of imprecision

“Guideline-Driven Medical Decision Limits”. HbA1c : Distribution for two samplings with set-point = cut-off Effect of Bias

“Guideline-Driven Medical Decision Limits”. Distribution of healthy set-points CV WITHIN-SUBJECT = 1.94 % HbA1c : Probability for a person with set-point = cut-off to be measured above cut-off

“Guideline-Driven Medical Decision Limits”. HbA1c : Distribution of set-points (healthy population) Cumulated frequency (probability functions) One sampling Increasing imprecision

“Guideline-Driven Medical Decision Limits”. HbA1c : Distribution of set-points (healthy population) Cumulated frequency (probability functions) One sampling Now we take a sample within a small interval of healthy set-points and multiply with the probability of these set- points exceed the cut-off to get the FP for healthy with this set-point and by repeating the process for all intervals we get the distribution of origins of set-points of FP

“Guideline-Driven Medical Decision Limits”. HbA1c : Varying imprecision and bias Bias = 0 % Bias = + 4 % The effect of positive bias is like moving the cut-off to the left Origin of set-points for healthy individuals measured above 48 mmol/mol (healthy diagnosed as diabetics)

“Guideline-Driven Medical Decision Limits”. HbA1c : Varying imprecision and bias Bias = 0 % Bias = + 4 % One Sampling Two Samplings One and two samplings One Sampling Origin of set-points for healthy individuals measured above 48 mmol/mol (healthy diagnosed as diabetics)

“Guideline-Driven Medical Decision Limits”. HbA1c : Percentage of healthy individuals measured > 48 mmol/mol (false positive diabetics) As function of bias % for varying percentages of imprecision For one sampling

“Guideline-Driven Medical Decision Limits”. HbA1c : As function of bias % for varying percentages of imprecision For one sampling For two samplings Percentage of healthy individuals measured > 48 mmol/mol false positive diabetics

HbA1c : What are the recommended quality specifications from What are the recommended quality specifications from Sacks et al. Clin Chem 2011;57:793-8 Desirable specifications for HbA1c measurement are an intralaboratory CV < 2% and an interlaboratory CV < 3.5 % The CV 3.5 % DCCT units corresponds to 5.2 % at 48 mmol/mol in IFCC units, and reduced by the 2 %, the final allowable bias is from ± 9 % at a 95 % interval and false positives could be from 0 to 2.8 % Personal information from Thomas Røraas and Sverre Sandberg, NOKLUS, Bergen, Norway

“Guideline-Driven Medical Decision Limits”. There is no reference interval for Cholesterol due to the strict decision limit of 6.2 mmol/L 95 % limits mg/dL = mmol/L Log-Gaussian distribution (natural logarithm) CV TOTAL =15.2 % ~ ln = CV WITHIN-SUBJECT = 6.0 % ~ ln = CV BETWEEN-SUBJECT = 13.9 % ~ ln = Recommended cut-off = 6.2 mmol/L ~ ln = Ricos et al. Scand J Clin Lab Invest 1999;59:491 Cholesterol: But a range for the total population can be estimated Klee et al. Scand J Clin Lab Invest 1999;59:509

“Guideline-Driven Medical Decision Limits”. Cholesterol : Distribution for a person with set-point = cut-off Effect of one and two samplings

“Guideline-Driven Medical Decision Limits”. Cholesterol : Distribution for persons with set-points below 6.2 mmol/L Cumulated frequency (probability functions) One sampling

“Guideline-Driven Medical Decision Limits”. Cholesterol : Distribution for persons with set-points below 6.2 mmol/L Cumulated frequency (probability functions) For one sampling For two samplings

“Guideline-Driven Medical Decision Limits”. Cholesterol : Distribution for persons with set-points below 6.2 mmol/L Cumulated frequency (probability functions), one sampling Now we take a sample within a small interval of healthy set-points and multiply with the probability of these set- points exceed the cut-off to get the FP for healthy with this set-point and by repeating the process for all intervals we get the distribution of origins of set-points of FP

“Guideline-Driven Medical Decision Limits”. Cholesterol : Origin of set-points Cumulated frequency (probability functions), one sampling One sampling Two samplings Bias = 0 %Bias = +4 % Apparent cut-off Assumed cut-off

“Guideline-Driven Medical Decision Limits”. Cholesterol : For one sampling For two samplings Percentage of false positive As function of bias % for varying percentages of imprecision

“Guideline-Driven Medical Decision Limits”. Biological Variables: HbA1c and Cholesterol Age and Gender Within- and Between-Subject Biological Variation Ethnicity, Lifestyle, Environment, Subclinical Diseases, etc. Seasonal variation CholesterolHbA1c Garde et al. Clin Chem 2000;46:551

1.Sharp decision limits are extremely sensitive to analytical bias 2.Use of two measurements in diagnosis reduces the effect of performance errors 3.Separate specifications for analytical bias and imprecision is recommended 4. Biological variation and other information should be considered in decision limits 5. A concentration related probability function as an alternative to sharp decision limit is proposed Conclusions “Guideline-Driven Medical Decision Limits”.

A concentration related probability function as an alternative to sharp decision limit “Guideline-Driven Medical Decision Limits”. Proposal : (Arbitrary Units) Probability of disease