july_research_ewald 5/8/04 9:49 AM Page 565

RESEARCH

Which test to detect micro- in diabetic patients? A systematic review

Ben Ewald, BMed, MMedSci, is a PhD scholar, Centre for Clinical Epidemiology and Biostatistics, University of Newcastle, and a general practitioner, Newcastle, New South Wales. John Attia, MD, PhD, FRCPC, is Senior Lecturer, Centre for Epidemiology and Biostatistics, University of Newcastle, New South Wales.

icroalbuminuria reflects early stages of The definition of used in BACKGROUND M , and is an independent prognostic and interventional research is Current guidelines suggest general 1 practitioners should screen their risk factor for cardiovascular disease. ‘albumin excretion of from 20 to 200 µg per diabetic patients for microalbuminuria. Structured care of diabetic patients such as minute demonstrated on two out of three There is a range of possible tests. that embodied in the Australian Common- timed samples’.3,4 Various thresholds We looked for studies that compared a wealth Government’s service incentive have been used for interpretation of ACR. timed urine sample (the gold standard) payment for includes annual screen- Older papers used cut points between 3.0 with a random spot sample. ing of all diabetic patients for micro- and 3.7 mg/mmol,5,6 while more recent METHOD albuminuria. However, there are differing rec- papers use a sex specific cut point of Systematic review and meta analysis ommendations about which test should be 2.5 mg/mmol in men and 3.5 mg/mmol in of studies comparing albumin to used. The gold standard for microalbuminuria women.7,8 Age and sex specific cut points for ratio (ACR) on a random is the measurement of ACR would increase specimen to albumin excretion rate albumin excretion rate ‘We suggest routine use of specificity in the Cardiab from an overnight or 24 hour timed (AER) on a 24 hour timed study.8 In the Cardiab sample. Studies were identified using ACR on a random urine Medline and EMBASE to June 2003. urine specimen. This may sample as the initial test in study, the prevalence of Studies were pooled using diagnostic be inconvenient for screening diabetics for albumin excretion in odds ratios and were checked for general practice patients microalbuminuria, thereby Australian diabetic as it requires collection of patients seen in general heterogeneity. saving the inconvenience of RESULTS all urine passed over the practice was 20% for collecting a timed urine speci- Ten studies covering 1470 patients specified time in a spe- microalbuminuria and 4% were included. Use of the ACR in cialised collection bag men with negligible loss of for macroalbuminuria.9 screening 100 diabetic patients would and generally two trips to case detection. One in eight The detection of con- miss only two out of the 20 patients a pathology centre. patients will require a timed firmed albuminuria or who would be expected to have Timed urine specimens specimen to clear up false microalbuminuria changes microalbuminuria, while there would are also prone to record- positives’. management in the fol- be 13 false positives. A timed ing errors, and lowing ways:10 the target specimen would be required to clarify the diagnosis for 31 patients. incomplete bladder emptying will bias for blood pressure treatment is lowered to 2 DISCUSSION results. A more convenient alternate test is 135/75, angiotensin converting enzyme (ACE) The marginal benefit of using a timed the albumin to creatinine ratio (ACR) done on inhibitors should be considered, renal function urine collection over a spot ACR to a single random urine sample that can be col- should be monitored and referral to a renal detect microalbuminuria in the lected at the time of a clinic visit. physician arranged if estimated creatinine screening of diabetic patients is small, Excretion of albumin fluctuates widely clearance falls below 30 mL/minute, although and not worth the cost and inconven- from day-to-day (and within any day), when targets for blood sugar levels and cardiovascu- ience of collecting a timed sample. standing (rather than lying), and greater with lar risk factors are unchanged efforts to modify activity (than rest); so any single estimate them may be increased, and, if the person may not be representative for that patient. does not have retinopathy, an alternate cause

Reprinted from Australian Family Physician Vol. 33, No. 7, July 2004565 july_research_ewald 5/8/04 9:49 AM Page 566

Research: Which test to detect microalbuminuria in diabetic patients?

for their renal disease should be sought. report a series of diabetic patients including off sensitivity against specificity, but the DOR We wondered what the marginal benefit adults, compare a random spot urine or first will remain nearly constant making it more was of a timed urine specimen over a spot morning UAC ratio with a timed overnight or suitable when combining studies. The DOR is specimen in the screening of diabetics for 24 hour urine albumin estimation, report sen- not clinically applicable, but can be used to micro- or macro-albuminuria in the general prac- sitivity and specificity (or provide sufficient derive summary sensitivity and specificity tice setting. detail so these values could be derived), and values. Heterogeneity of the DORs was include more than 20 subjects. tested using the Breslow-Day Q test, and a Method Several studies reported ACR on urine summary DOR calculated by the random We searched Medline and EMBASE data- taken from the 24 hour collection bag. effects method of DerSimonian and Laird in bases from 1966 to February 2004 using the However, as this does not reflect the varia- Stats Direct.13 MeSH terms ‘diabetes mellitus’, ‘’, tion likely to occur with a random specimen, ‘albuminuria’, ‘area under curve’, and ‘sensi- these studies were not included.3 As these Results tivity and specificity’. Text searches were criteria served as a quality filter, no further We identified 43 studies, of which 12 fitted also conducted for these terms as well as quality scoring was necessary. the retrieval criteria. Two were excluded: one albumin, creatinine, microalbumin, ACR and reported multiple samples per patient,14 Analysis urine albumin to creatinine (UAC). another reported nondiabetic patients,4 References of retrieved papers were also The diagnostic odds ratio (DOR) was calcu- leaving 10 remaining studies (Table 1). There searched for eligible studies. Retrieval criteria lated for each study: it is the product in the was heterogeneity between the 10 studies were developed with regard to the STARD11 2x2 table of true tests divided by the product (p<0.001). However, when analysis was statement of reporting diagnostic research. of false (a x d/b x c).12 restricted to the seven studies that used a 24 However, few papers met all the require- hour AER rather than an overnight reference ments of STARD so it could not be used as True positives (a) False positives (b) standard, they were found to be homoge- inclusion criteria. Papers were of acceptable nous (p=0.44) (Figure 1). Neither the use of a quality if they set out to compare diagnostic False negatives (c) True negatives (d) sex specific cut point for ACR, or the setting performance of the two tests, all subjects (general practice versus hospital clinic) sys- had both tests, cut points were described, Its benefit is that it is not influenced by the tematically affected the results. and sufficient detail was presented to meet choice of cut point used in the various The summary odds ratio estimated with a our inclusion criteria as follows. They had to studies. Changes to the cut point will trade random effects model was 45.8 (95% CI:

Table 1. Studies comparing random ACR to a timed AER specimen

Study Setting Comparison Sensitivity Specificity Prevalence of albuminuria (%)

Zelmanovitz5 54, hospital outpatient clinic Brazil Random ACR vs 24 hour AER 89 89 81 Wiegman6 135, USA hospital outpatient clinic Random ACR vs 24 hour AER 82 81 22 90 with type 1 diabetes Gatling15 311, 40 UK GPs Random ACR vs timed overnight AER 80 81 6 Jermendy16 192, Hungarian diabetes centre Morning ACR vs timed overnight AER 75 69 68 Claudi7 106, Norwegian primary care Random ACR on near patient analyser 90 90 NA vs overnight AER Ahn17 105, Korean ambulatory patients Random ACR vs 24 hour AER 77 92 52 Eshoj18 54, type 1 Denmark Morning urine vs 24 hour AER 90 88 46 Ng19 65, Singapore outpatients Morning ACR on desk top DCA2000 71 98 22 vs 24 hour AER Gyamlani20 136, adult diabetics, Early morning ACR vs 24 hour AER 85 85 29 Olmsted county USA Houlihan8 314, Australian hospital outpatients Spot morning sample vs 24 hour AER 95 81 Male 54% Female 32%

566 Reprinted from Australian Family Physician Vol. 33, No. 7, July 2004 july_research_ewald 5/8/04 9:49 AM Page 567

Research: Which test to detect microalbuminuria in diabetic patients?

would also incorrectly identify 13 patients as Zelmanovitch 70.20 (6.48, 3149.11) having albuminuria, requiring them to Ahn 37.15 (10.29, 161.08) Gyamlani 30.07 (9.88, 99.85) proceed to albumin measurement on a timed Eshoj 88.67 (9.38 3807.82) urine sample. The use of an ACR followed by Weigman 21.25 (6.70, 77.60) a confirmatory timed AER would result in Ng 125.00 (11.05, 5519.52) Houlihan 81.04 (33.26, 219.97) 31 patients requiring a timed urine collection 24 hour subgroup 45.76 (28.54, 73.38) compared to 100 patients undergoing this Heterog p=0.44 Gatling 16.79 (5.09, 70.79) test if it was used initially. Claudi 87.88 (20.38, 425.80) Insisting on a timed urine collection may Jermendy 6.56 (3.20, 13.61) reduce adherence to the screening regimen. 12 hour subgroup If the inconvenience of a timed specimen Heterog p=0.0001 19.84 (5.44–72.34) Pooled DOR resulted in 10 of the 100 patients not being Heterog p=0.001 35.53 (16.31, 77.40) screened at all, two patients with microalbu- minuria would be missed, completely 0.2 0.5 1 25 10 100 1000 1.00E+05 cancelling the benefit of the better test. Diagnostic odds ratio (95% confidence interval) As screening for albuminuria is imple-

Figure 1. Diagnostic odds ratios of included studies by reference standard and summary estimate mented in a practice, the initial round can be expected to find the problem in around 24% 28.5–73.4). The odds ratio can be translated an analysis of sensitivity and specificity of patients. However, subsequent rounds of to a pair of sensitivity and specificity values against the ‘two out of three’ definition.16 screening will find only new cases that have by choosing the sensitivity value from one of In another study, the prevalence of developed in the intervening year. If these the larger studies and calculating the speci- microalbuminuria in diabetics was 31% in incident cases of microalbuminuria occur in ficity based on the odds ratio by solving one of three tests, but only 20% when a ‘two 2% of diabetics per year, the use of ACR simultaneous equations. Sensitivity of 90% out of three’ definition was used.2 The vari- instead of a timed specimen will miss 0.2 as in the study by Claudi,7 gives specificity of ability of the gold standard means the of a case per 100 diabetics screened, and 84% (95% CI: 76–89) or if sensitivity was inaccuracy of the ACR was overestimated. 15 patients will need to proceed to a timed 80% as in Gatling,15 specificity would be 92% Two other possible screening tests have specimen to clear up false positives. (95% CI: 88–95). not been assessed in this review. The deter- We suggest routine use of ACR on a random mination of albumin concentration alone on a urine sample as the initial test in screening dia- Discussion random specimen does not perform as well betics for microalbuminuria, thereby saving the Research into diagnostic processes relies on as the ACR3 and uses the same specimen. inconvenience of collecting a timed urine speci- a gold standard of diagnosis against which High sensitivity albumin test strips such as men with negligible loss of case detection. One other tests are compared. Discordance of ‘Micral’ are generally not used in Australian in eight patients will require a timed specimen test results is interpreted as error in the can- general practice for administrative reasons: to clear up false positives. didate test and assumes the reference test is the cost of $1.80 per test strip would be always correct. Such research is therefore borne by the practice while the cost of any Conflict of interest: none declared. only as good as the gold standard. The laboratory based test is borne by Medicare. studies we reviewed reported spot urine References How should we apply these results 1. Dinneen SF, Gerstein HC. The association of tests against a single timed specimen, which in practice? microalbuminuria and mortality in noninsulin does not match the ‘two out of three’ defini- dependent diabetes mellitus. A systematic tion of microalbuminuria used in the literature Applying a sensitivity of 90% and specificity overview of the literature. Arch Intern Med for prognostic and intervention studies. of 84% to the screening of 100 diabetic 1997;157:1413–1418. 2. Ellis D, Coonrod BA, Dorman JS, et al. Choice of In one study, both timed albumin excretion patients, of whom 20 actually had microalbu- urine sample predictive of microalbuminuria in and ACR were measured sequentially three minuria, the use of a random ACR instead of patients with insulin dependent diabetes melli- times on each subject to categorise them as a timed overnight or 24 hour sample would tus. Am J Dis 1989;13:321–328. 3. Bakker AJ. Detection of microalbuminuria. normal, microalbuminuria, or macroalbumin- result in the problem being missed at that Receiver operating characteristic curve analysis uria.16 The timed specimen gave consistent screening round in two patients. Based on favors albumin-to-creatinine ratio over albumin categorisation 80% of the time, while ACR Australian data, a further four patients would concentration. Diabetes Care 1999;22:307–313. 4. Morgensen C, Vestbo E, Poulsen P, Christiansen gave consistent categorisation 75% of the have macroalbuminuria which either test C. Microalbuminuria and potential confounders. time. Unfortunately this paper did not present would detect. The use of the simpler test Diabetes Care 1995;18:572–581.

Reprinted from Australian Family Physician Vol. 33, No. 7, July 2004 567 july_research_ewald 5/8/04 9:49 AM Page 568

Research: Which test to detect microalbuminuria in diabetic patients?

5. Zelmanovitz T, Gross J, Oliveira J. The receiver operating characteristics curve in the evaluation of a random urine specimen as a screening test for diabetic nephropathy. Diabetes Care 1997;20:516–519. 6. Wiegmann T, Chonko A, Barnard M. Comparison of albumin excretion rate obtained with different times of collection. Diabetes Care 1990;13:864–871. 7. Claudi T, Cooper J. Comparison of urinary albumin excretion in overnight urine and albumin creatinine ratio in spot urine in dia- betic patients in general practice. Scand J Primary Health Care 2001;19:247–248. 8. Houlihan C, Tsalamandris C. Albumin to creati- nine ratio: a screening test with limitations. Am J Kidney Dis 2002;39:1–10. 9. Carter S, Burns J, Bonney M. CARDIAB. Sydney: Centre for general practice integration studies, University of NSW, 2000. 10. McIntosh A. RCGP Clinical guideline for . Diabetic renal disease: prevention and early management. Sheffield: Royal College of General Practitioners, 2002. 11. Bossuyt PRJ. The STARD statement for reporting studies of diagnostic accuracy: explanation and elaboration. Clin Chem 2003;49:7–18. 12. Deeks JJ. Systematic reviews in health care: sys- tematic reviews of evaluations of diagnostic and screening tests. BMJ 2001;323:157–162. 13. Stats Direct L. Stats Direct. In: 2.2.8 edn. Cheshire UK: Stats Direct, 2003. 14. Nathan DM, Rosenbaum C, Protasowicki VD. Single void urine samples can be used to esti- mate quantitative microalbuminuria. Diabetes Care. 1987;10:414–418. 15. Gatling W, Knight C, Mullee MA, Hill RD. Microalbuminuria in diabetes: a population study of the prevalence and an assessment of three screening tests. Diabet Med 1988;5:343–347. 16. Jermendy G, Farkas K, Nadas J, Daroczy A, Peterfai E. Practical aspects of measuring microalbuminuria in diabetic patients. Diabetes Nutr Metab Clin Exper 2001;14:195–200. 17. Ahn CW, Song YD, Kim JH, et al. The validity of random urine specimen albumin measurement as a screening test for diabetic nephropathy. Yonsei Medical Journal 1999;40:40–45. 18. Eshoj O, Feldt-Rasmussen B, Larsen ML, Mogensen EF. Comparison of overnight, morning and 24 hour urine collections in the assessment of diabetic microalbuminuria. Diabet Med 1987;4:531–533. 19. Ng WY, Lui KF, Thai AC. Evaluation of a rapid screening test for microalbuminuria with a spot measurement of urine albumin-creatinine ratio. Ann Acad Med Singapore 2000;29:62–65. 20. Gyamlani G, Bergstralh E, Slezak J. Urinary albumin to osmolality ratio predicts 24 hour urine albumin excretion in diabetes mellitus. Am J Kidney Dis 2003;42:685–692. AFP

Correspondence Email: [email protected]

568 Reprinted from Australian Family Physician Vol. 33, No. 7, July 2004