Volume 56, Issue 4

This review focuses on machine learning and on how methods and models combining data analytics and artificial intelligence have been applied to laboratory medicine so far. Although still in its infancy, the potential for applying machine learning to laboratory data for both diagnostic and prognostic purposes deserves more attention by the readership of this journal, as well as by physician-scientists who will want to take advantage of this new computer-based support in pathology and laboratory medicine.

Irisin is produced by a proteolytic cleavage of fibronectin type III domain-containing protein 5 (FNDC5) and has emerged as a potential mediator of exercise-induced energy metabolism. The purpose of this study was to review the results of studies that investigated irisin responses to acute and chronic exercise and provide an update. A comprehensive search in the databases of MEDLINE was performed (74 exercise studies). The focus of the analysis was on data concerning FNDC5 mRNA expression in skeletal muscle and circulating irisin concentration relatively to exercise mode, intensity, frequency and duration and the characteristics of the sample used. Circulating irisin levels may either not relate to FNDC5 transcription or expression of the later precedes irisin rise in the blood. Acute speed/strength and endurance exercise protocols represent potent stimuli for irisin release if they are characterized by adequate intensity and/or duration. There are no reports regarding irisin responses to field sport activities. Although animal studies suggest that irisin may also respond to systematic exercise training, the majority of human studies has produced contradictory results. Certain methodological issues need to be considered here such as the analytical assays used to measure irisin concentration in the circulation. Results may also be affected by subjects’ age, conditioning status and exercise intensity. The role of irisin as a moderator of energy metabolism during exercise remains to be seen.

Internal Quality Control and External Quality Assurance are separate but related processes that have developed independently in laboratory medicine over many years. They have different sample frequencies, statistical interpretations and immediacy. Both processes have evolved absorbing new understandings of the concept of laboratory error, sample material matrix and assay capability. However, we do not believe at the coalface that either process has led to much improvement in patient outcomes recently. It is the increasing reliability and automation of analytical platforms along with improved stability of reagents that has reduced systematic and random error, which in turn has minimised the risk of running less frequent IQC. We suggest that it is time to rethink the role of both these processes and unite them into a single approach using an Average of Normals model supported by more frequent External Quality Assurance samples. This new paradigm may lead to less confusion for laboratory staff and quicker responses to and identification of out of control situations.

Background: Uridine diphosphate glycosyltransferases 1A1 (UGT1A1) plays an essential role in detoxification and excretion of several endogenous and exogenous compounds. A functional polymorphism in the promoter of the UGT1A1 gene (TA repeat insertion, UGT1A1 *28, rs3064744) has been associated with reduced UGT1A1 enzyme activity. The purpose of the present study was to investigate the role of UGT1A1 genotypes in mortality. Methods: UGT1A1 genotypes as well as baseline plasma bilirubin levels were analyzed in participants of the Ludwigshafen Risk and Cardiovascular Health study (n=3316). UGT1A1 *28 genotypes were determined on an ABI PRISM 3730 genetic analyzer. Results: As expected, UGT1A1 genotypes were associated with baseline bilirubin levels (*1/*1 genotype: 9.1±4.6 µmol/L; *1/*28 genotype: 10.8±5.3; *28/*28: 16.9±9.2; p<0.001). During a median follow-up of 10.4 years, 995 subjects (30.0%) died. In a multivariate regression analysis adjusting for age, sex, smoking, type 2 diabetes, dyslipidemia, alanine aminotransferase (ALT) levels and bilirubin levels, the UGT1A1 *28 variant predicted lower overall mortality (hazard ratio [HR], 0.86; 95% confidence interval [CI], 0.78–0.95; p=0.003). Contrary to expected, higher baseline bilirubin levels predicted increased mortality (HR, 1.014; 95% CI, 1.002–1.025; p=0.019). Conclusions: The UGT1A1 *28 gene variant is associated with lower mortality rates. The protective effect of the UGT1A1 *28 variant likely includes mechanism other than bilirubin metabolism.

Background: There are several complementary English-language guidelines for the performance of the sweat chloride test. These guidelines also incorporate information for the collection of conductivity samples. However, recommendations for the measurement and reporting of sweat conductivity are less clear than for sweat chloride. The aim of the study was to develop an understanding of the testing and reporting practices of sweat conductivity in Australasian laboratories. Methods: A survey specifically directed at conductivity testing was sent to the 12 laboratories registered with the Royal College of Pathologists of Australasia Quality Assurance Programs. Results: Nine (75%) laboratories participated in the survey, seven of whom used Wescor Macroduct ® for collecting sweat and the Wescor SWEAT·CHEK™ for conductivity testing, and the remaining two used the Wescor Nanoduct ® . There was considerable variation in frequency and staffing for this test. Likewise, criteria about which patients it was inappropriate to test, definitions of adequate collection sweat rate, cutoffs and actions recommended on the basis of the result showed variations between laboratories. Conclusions: Variations in sweat conductivity testing and reporting reflect many of the same issues that were revealed in sweat chloride test audits and have the potential to lead to uncertainty about the result and the proper action in response to the result. We recommend that sweat testing guidelines should include clearer statements about the use of sweat conductivity.

Background: The key incident monitoring and management systems (KIMMS) quality assurance program monitors incidents in the pre- and postanalytical phases of testing in medical laboratories. Haemolysed specimens have been found to be the most frequent preanalytical error and have major implications for patient care. The aims of this study were to assess the suitability of KIMMS for quality reporting of haemolysis and to devise a meaningful method for reporting and monitoring haemolysis. Methods: A structured survey of 68 Australian KIMMS laboratory participant organisations was undertaken. Quarterly haemolysis reports (2011–2014) were analysed. Results: Among 110 million accessions reported, haemolysis rates varied according to the reporting methods that participants used for assigning accessions (16% of participants reported haemolysis by specimen and 83% reported by episode) and counting haemolysis rejections (61% by specimen, 35% by episode and 3% by test). More than half of the participants (56%) assigned accessions by episode and counted rejections by specimen. For this group, the average haemolysis rate per 100,000 episodes was 177 rejected specimens with the average rate varying from 100 to 233 over time. The majority of participants (91%) determined rejections using the haemolysis index. Two thirds of participants (66%) recorded the haemolysis manually in laboratory information systems. Conclusions: KIMMS maintains the largest longitudinal haemolysis database in the world. However, as a means of advancing improvements in the quality of the preanalytical laboratory process, there is a need to standardise reporting methods to enable robust comparison of haemolysis rejection rates across participant laboratories.

Background: Manual handling of hemolyzed samples is not standardized and is vulnerable to errors. This study aimed to evaluate laboratory errors due to manual handling of hemolyzed samples and to assess the risk they might have for patient safety. Methods: Data were retrospectively obtained from a laboratory information system for 25 emergency tests from hemolyzed samples. Hemolysis (concentration of free hemoglobin >0.5 g/L) was visually assessed by comparison with a color chart. The reference person reestimated the routinely assessed degree of hemolysis to all samples (n=3185) received in the laboratory in a 1-week period. For each test, the correct and incorrect way of handling results was determined. Risk assessment was performed according to ISO 14971 standard with five categories of risk (S1–S5) and error occurrence (O1–O5). Results: In the studied period, the emergency laboratory received 495 hemolyzed samples (15.5%) with a total of 2518 laboratory test requests (15.5%): 102 (20.6%) of the reports from hemolyzed samples had a comment on hemolysis; 31% of the test results were handled incorrectly (20.7% due to the incorrect release of the test result despite hemolysis interference and 10.3% due to unnecessary suppression), accounting for 4.8% of the total test volume. Tests with the highest combination of risk and occurrence rate were troponin T, potassium and total bilirubin. Conclusions: Manual handling of hemolyzed samples may lead to risk of errors in reporting results for troponin T, potassium and total bilirubin, which may have an effect on clinical decision. In addition, unnecessary suppression of the sample results unaffected by hemolysis could affect patient outcome.

Background A number of factors regarding blood collection, handling and storage may affect sample quality. The purpose of this study was to assess the impact on plasma protein profiles by delayed centrifugation and plasma separation and multiple freeze-thaw cycles. Methods Blood samples drawn from 16 healthy individuals were collected into ethylenediaminetetraacetic acid tubes and kept either at 4 °C or 22 °C for 1–36 h prior to centrifugation. Plasma samples prepared 1 h after venipuncture were also subjected to two to eight cycles of freezing at −80 °C and thawing at 22 °C. Multiplex proximity extension assay, an antibody-based protein assay, was used to investigate the influence on plasma proteins. Results Up to 36 h delay before blood centrifugation resulted in significant increases of 16 and 40 out of 139 detectable proteins in samples kept at 4 °C or 22 °C, respectively. Some increases became noticeable after 8 h delay at 4 °C but already after 1 h at 22 °C. For samples stored at 4 °C, epidermal growth factor (EGF), NF-kappa-B essential modulator, SRC, interleukin 16 and CD6 increased the most, whereas the five most significantly increased proteins after storage at 22 °C were CD40 antigen ligand (CD40-L), EGF, platelet-derived growth factor subunit B, C-X-C motif chemokine ligand 5 and matrix metallopeptidase 1 (MMP1). Only matrix metallopeptidase 7 (MMP7) decreased significantly over time and only after storage at 22 °C. No protein levels were found to be significantly affected by up to eight freeze-thaw cycles. Conclusions Plasma should be prepared from blood after a limited precentrifugation delay at a refrigerated temperature. By contrast, the influence by several freeze-thaw cycles on detectable protein levels in plasma was negligible.

Background: The International Federation of Clinical Chemistry and Laboratory Medicine has introduced in recent times the turnaround time (TAT) as mandatory quality indicator for the postanalytical phase. Classic TAT indicators, namely, average, median, 90th percentile and proportion of acceptable test (PAT), are in use since almost 40 years and to date represent the mainstay for gauging the laboratory timeliness. In this study, we investigated the performance of the Six Sigma Z-score, which was previously introduced as a device for the quantitative assessment of timeliness. Methods: A numerical simulation was obtained modeling the actual TAT data set using the log-logistic probability density function. Five thousand replicates for each size of the artificial TAT random sample (n=20, 50, 250 and 1000) were generated, and different laboratory conditions were simulated manipulating the PDF in order to generate more or less variable data. The Z-score and the classic TAT indicators were assessed for precision (%CV), robustness toward right-tailing (precision at different sample variability), sensitivity and specificity. Results: Z-score showed sensitivity and specificity comparable to PAT (≈80% with n≥250), but superior precision that ranged within 20% by moderately small sized samples (n≥50); furthermore, Z-score was less affected by the value of the cutoff used for setting the acceptable TAT, as well as by the sample variability that reflected into the magnitude of right-tailing. Conclusions: The Z-score was a valid indicator of laboratory timeliness and a suitable device to improve as well as to maintain the achieved quality level.

Background: There is no external quality assessment (EQA) program for hemoglobin analysis that uses lyophilized hemoglobin control materials with HbA 2 /E in levels as high as those found in people with the β-thalassemia trait, HbE trait, β-thalassemia/HbE disease or homozygous HbE; these are all found frequently in the southeast Asian population. The aim of this study was to evaluate the efficiency of the control materials used in the established proficiency testing (PT) program at the Associated Medical Sciences-Clinical Service Center (AMC-CSC), Chiang Mai University, Chiang Mai, Thailand. Methods: The PT program for Hb analysis and the thalassemia interpretation was established in compliance with ISO/IEC17043:2010. Three cycles per year were performed in 2015 and 2016. In each cycle, three different types of control material were provided to the participants. Each participant analyzed the control materials in the same manner as in their routine practices. Hb analysis results and their thalassemia interpretation codes were entered into the report form and sent back to AMC-CSC. Results: The number of participants increased from 63 in 2015 to 76 in 2016. In addition, the number of participants who took part in all three cycles increased from 95.2% (60/63) in 2015 to 100% (76/76) in 2016. All participants reported the correct Hb measurement and type; however, misinterpretations in thalassemia diagnosis were noted. Conclusions: The lyophilized hemoglobin control materials prepared at AMC-CSC were used successfully in our PT program. However, the study results indicate the need for further improvement in thalassemia interpretation skills for laboratory staff.

Background: The determination of κ free light chains (KFLC) in cerebrospinal fluid (CSF) by nephelometry is a feasible alternative to immunoglobulin G oligoclonal bands (OCB) in the evaluation of intrathecal synthesis of immunoglobulin in multiple sclerosis (MS) and other demyelinating diseases. The aim of this study was to assess the diagnostic value of KFLC and its inclusion in a procedure algorithm along with OCB interpretation. Methods: A cross-sectional study, which included 123 patients with a CSF OCB request, was carried out. Isoelectric focusing followed by immunofixation was used to detect OCB, and nephelometry was used to analyze KFLC. The KFLC index was calculated using CSF/serum quotient of KFLC and albumin. The KFLC index was compared with MS diagnosis to find the optimal cutoff. It was obtained from the receiver operating characteristic (ROC) curves and the Youden method. Results: The CSF KFLC median was 1.66 mg/L in the MS group, whereas in other central nervous system diseases, KFLC showed generally no or only moderate increase in CSF (median 0.10 mg/L). KFLC index showed a significant difference between groups. ROC analysis for CSF KFLC concentration, and KFLC indexes were 91.88% and 93.94%, respectively. The best cutoff for the KFLC index was 2.91 for MS diagnosis (sensitivity: 83.78%; specificity: 85.88%). The proposed algorithm showed high sensitivity (89.19%) and specificity (84.71%). Conclusions: KFLC determination is rapid and automatized, but it has no higher sensitivity and specificity than OCB in MS diagnosis. Nevertheless, when used in screening, it could reduce the number of manual OCB tests.

Background: The aim of the study was to determine the prevalence and clinical associations of antiphosphatidylserine/prothrombin antibodies (aPS/PT) with thrombosis and pregnancy loss in Chinese patients with antiphospholipid syndrome (APS) and seronegative APS (SNAPS). Methods: One hundred and eighty six Chinese patients with APS (67 primary, 119 secondary), 48 with SNAPS, 176 disease controls (79 systemic lupus erythematosus [SLE], 29 Sjogren’s syndrome [SS], 30 ankylosing spondylitis [AS], 38 rheumatoid arthritis [RA]) and 90 healthy donors were examined. IgG and IgM aPS/PT, IgG/IgM/IgA anticardiolipin (aCL) and IgG/IgM/IgA anti-β 2 -glycoprotein I (anti-β2GPI) antibodies were tested by ELISA. Results: One hundred and sixty (86.0%) of APS patients were positive for at least one aPS/PT isotype. One hundred and thirty five (72.6%) were positive for IgG aPS/PT, 124/186 (66.7%) positive for IgM aPS/PT and 99 (53.2%) positive for both. Approximately half of the SNAPS patients were positive for IgG and/or IgM aPS/PT. Highly significant associations between IgG aPS/PT and venous thrombotic events (odds ratio [OR]=6.72) and IgG/IgM aPS/PT and pregnancy loss (OR=9.44) were found. Levels of IgM aPS/PT were significantly different in APS patients with thrombotic manifestations and those with fetal loss (p=0.014). The association between IgG/IgM aPS/PT and lupus anticoagulant (LAC) was highly significant (p<0.001). When both were positive, the OR for APS was 101.6. Notably, 91.95% (80/87) of LAC-positive specimens were positive for IgG and/or IgM aPS/PT, suggesting aPS/PT is an effective option when LAC testing is not available. Conclusions: Anti-PS/PT antibody assays demonstrated high diagnostic performance for Chinese patients with APS, detected some APS patients negative for criteria markers and may serve as potential risk predictors for venous thrombosis and obstetric complications.

Background: An evaluation of SENTiFIT ® 270 (Sentinel Diagnostics, Italy; Sysmex, Spain) analyser for the quantitation of faecal haemoglobin (f-Hb) was performed. Methods: The analytical imprecision, linearity, carry over and f-Hb stability were determined. Evaluation of the diagnostic accuracy was performed on 487 patients. Results: Within-run and between-run imprecision ranged 1.7%–5.1% and 3.8%–6.2%, respectively. Linearity studies revealed a mean recovery of 101.1% (standard deviation, 6.7%) for all dilutions. No carry over was detected below 7650 μg Hb/g faeces. Decay of f-Hb in refrigerated samples ranged 0.2%–0.5% per day. f-Hb in patients with advanced colorectal neoplasia (ACRN) (colorectal cancer [CRC] plus advanced adenoma [AA]) were significantly higher than from those with a normal colonoscopy. Sensitivity for ACRN at f-Hb cutoffs from 10 to 60 μg Hb/g faeces ranged from 28.9% (95% confidence interval [CI], 21.7%–37.2%) to 46.5% (95% CI, 38.1%–55%), the specificity ranged from 85% (95% CI, 82.3%–87.3%) to 93.2% (95% CI, 91.2%–94.8%), positive predictive values for detecting CRC and AA ranged from 11.6% (95% CI, 7.6%–17.2%) to 20.6% (95% CI, 13.3%–30.3%) and from 34.7% (95% CI, 28.1%–42%) to 42.3% (95% CI, 32.4%–52.7%), respectively, and the negative predictive value for ACRN ranged from 90.2% (95% CI, 87.9%–92.2%) to 88.4% (95% CI, 86%–90.4%). Using two samples per patient sensitivity increased with a slight decrease in specificity. Conclusions: The analytical and clinical performances of SENTiFIT assay demonstrate a specific and accurate test for detecting ACRN in symptomatic patients and those undergoing surveillance.

Background: Proper management of patients with chronic hepatitis B virus (HBV) infection requires monitoring of plasma or serum HBV DNA levels using a highly sensitive nucleic acid amplification test. Because commercially available assays differ in performance, we compared herein the performance of the Hologic Aptima HBV Quant assay (Aptima) to that of the Roche Cobas TaqMan HBV test for use with the high pure system (HPS/CTM). Methods: Assay performance was assessed using HBV reference panels as well as plasma and serum samples from chronically HBV-infected patients. Method correlation, analytical sensitivity, precision/reproducibility, linearity, bias and influence of genotype were evaluated. Data analysis was performed using linear regression, Deming correlation analysis and Bland-Altman analysis. Results: Agreement between the assays for the two reference panels was good, with a difference in assay values vs. target <0.5 log. Qualitative assay results for 159 clinical samples showed good concordance (88.1%; κ=0.75; 95% confidence interval: 0.651–0.845). For the 106 samples quantitated by both assays, viral load results were highly correlated (R=0.92) and differed on average by 0.09 log, with 95.3% of the samples being within the 95% limit of agreement of the assays. Linearity for viral loads 1–7 log was excellent for both assays (R 2 >0.98). The two assays had similar bias and precision across the different genotypes tested at low viral loads (25–1000 IU/mL). Conclusions: Aptima has a performance comparable with that of HPS/CTM, making it suitable for use for HBV infection monitoring. Aptima runs on a fully automated platform (the Panther system) and therefore offers a significantly improved workflow compared with HPS/CTM.

Background: The aim of the study was to establish reference intervals for 24-h urinary stone risk factors in the healthy Chinese Han population. Methods: From May 2013 to July 2014, we collected and analyzed 24-h urine samples from healthy adult Han population during a cross-sectional study across China. The protocol for analysis of 24-h urine included volume, pH, oxalate, citrate, sodium, potassium, chloride, calcium, phosphorous, creatinine, urate, magnesium, the ion activity products of calcium oxalate (AP(CaOx) index s ) and calcium phosphate (AP(CaP) index s ). We calculated the reference intervals according to the Clinical and Laboratory Standards Institute (CLSI) 2008 guidelines and compared them with those recorded in other studies. Results: A total of 132 male and 123 female healthy subjects with a mean (SD, range) age of 52.4 (15.2, 19–89) years were eligible in the final analysis. Men had higher 24-h excretion of creatinine, calcium, urate and phosphorus and lower levels of citrate, magnesium, chloride, sodium and potassium than women. AP(CaOx) index s and AP(CaP) index s were significantly higher among men than women. When urinary findings were compared with the reference intervals, most of our data showed a high abnormality rate, especially for creatinine, calcium, citrate, magnesium, chloride, sodium and potassium. Conclusions: The present study revealed the normal metabolic status for stone risk factors of the Chinese Han population. It is therefore necessary for each country or region to define their own reference intervals for comparison of stone risk factors between patients and healthy subjects.

Background: Point of care (POC) assays for cardiac troponins I or T (cTnI or cTnT) may accelerate the diagnosis of patients with suspected acute coronary syndrome (ACS). However, their clinical utility according to the 0 h/3 h algorithm recommended by the European Society of Cardiology (ESC) for non-ST elevation myocardial infarction (NSTEMI) is unknown. Methods: Blood samples from 90 patients with suspected ACS were obtained at hospital admission and 3 h later. Concentrations of cTn were determined using five POC assays (AQT90 FLEX cTnI and cTnT; PATHFAST™ cTnI; Stratus CS 200 cTnI; and Triage MeterPro cTnI) and two guideline-acceptable high-sensitivity (hs) immunoassays. Results: For the diagnosis of NSTEMI (n=15), AUCs for Abbott hs-cTnI and Roche hs-cTnT were 0.86 [95% confidence interval (CI), 0.75–0.96] and 0.88 (95% CI, 0.80–0.95), respectively, at admission, and 0.96 and 0.94, respectively, 3 h later. With the 99th percentile cutoff, their sensitivities were 62% and 92%, respectively, at admission, and 77% and 100%, respectively, 3 h later. The PATHFAST™ cTnI assay showed AUCs of 0.90 (95% CI, 0.82–0.97) and 0.94 (95% CI, 0.89–1.00), respectively, and sensitivities of 67% and 75% at admission and 3 h later, respectively. The other cTn POC assays had AUCs of 0.71 (95% CI, 0.53–0.89) to 0.84 (95% CI, 0.71–0.96) and 0.86 (95% CI, 0.72–0.99) to 0.87 (95% CI, 0.75–0.99) and sensitivities of 39%–50% and 62%–77% at admission and 3 h later, respectively. Conclusions: PATHFAST™ cTnI assay proved itself as comparable to ESC-guideline acceptable hs-cTn assays. The lower sensitivity of the other POC assays limits their clinical utility and would require longer follow-up monitoring of patients for the safe NSTEMI rule-out.

Background: Inflammatory biomarkers may aid to distinguish between systemic inflammatory response syndrome (SIRS) vs. sepsis. We tested the hypotheses that (1) presepsin, a novel biomarker, can distinguish between SIRS and sepsis, and (2) higher presepsin levels will be associated with increased severity of illness and (3) with 28-day mortality, outperforming traditional biomarkers. Methods: Procalcitonin (PCT), C-reactive protein (CRP), presepsin, and lactate were analyzed in 60 consecutive patients (sepsis and SIRS, n=30 per group) on day 1 (D1) to D3 (onset sepsis, or after cardiac surgery). The systemic organ failure assessment (SOFA) score was determined daily. Results: There was no difference in mortality in sepsis vs. SIRS (12/30 vs. 8/30). Patients with sepsis had higher SOFA score vs. patients with SIRS (11±4 vs. 8±5; p=0.023), higher presepsin (AUC=0.674; p<0.021), PCT (AUC=0.791; p<0.001), CRP (AUC=0.903; p<0.0001), but not lactate (AUC=0.506; p=0.941). Unlike other biomarkers, presepsin did not correlate with SOFA on D1. All biomarkers were associated with mortality on D1: presepsin (AUC=0.734; p=0.0006; best cutoff=1843 pg/mL), PCT (AUC=0.844; p<0.0001), CRP (AUC=0.701; p=0.0048), and lactate (AUC=0.778; p<0.0001). Multiple regression analyses showed independent associations of CRP with diagnosis of sepsis, and CRP and lactate with mortality. Increased neutrophils (p=0.002) and decreased lymphocytes (p=0.007) and monocytes (p=0.046) were also associated with mortality. Conclusions: Presepsin did not outperform traditional sepsis biomarkers in diagnosing sepsis from SIRS and in prognostication of mortality in critically ill patients. Presepsin may have a limited adjunct value for both diagnosis and an early risk stratification, performing independently of clinical illness severity.

Background: The release of hormones from the adrenal gland is vital in acute and chronic illnesses such as chronic obstructive pulmonary disease (COPD) involving recurrent exacerbations. Using a metabolomic approach, we aim to investigate associations of different adrenal hormone metabolites with short- and long-term mortality in COPD patients. Methods: We prospectively followed 172 COPD patients (median age 75 years, 62% male) from a previous Swiss multicenter trial. At baseline, we measured levels of a comprehensive spectrum of adrenal hormone metabolites, including glucocorticoid, mineralocorticoid and androgen hormones by liquid chromatography coupled with tandem mass spectrometry (MS). We calculated Cox regression models adjusted for gender, age, comorbidities and previous corticosteroid therapy. Results: Mortality was 6.4% after 30 days and increased to 61.6% after 6 years. Higher initial androgen hormones predicted lower long-term mortality with significant results for dehydroepiandrosterone (DHEA) [adjusted hazard ratio (HR), 0.82; 95% confidence interval (CI), 0.70–0.98; p=0.026] and dehydroepiandrosterone sulfate (DHEA-S) (adjusted HR, 0.68; 95% CI, 0.50–0.91; p=0.009). An activation of stress hormones (particularly cortisol and cortisone) showed a time-dependent effect with higher levels pointing towards higher mortality at short term, but lower mortality at long term. Activation of the mineralocorticoid axis tended to be associated with increased short-term mortality (adjusted HR of aldosterone, 2.76; 95% CI, 0.79–9.65; p=0.111). Conclusions: Independent of age, gender, corticosteroid exposure and exacerbation type, adrenal hormones are associated with mortality at short and long term in patients with COPD exacerbation with different time-dependent effects of glucocorticoids, androgens and mineralocorticoids. A better physiopathological understanding of the causality of these effects may have therapeutic implications.