- Email: [email protected]
Monitoring young people with type 1 diabetes for diabetic nephropathy: Potential errors of annual ACR testing
diabetes research and clinical practice 99 (2013) 307–314
Contents available at Sciverse ScienceDirect
Diabetes Research and Clinical Practice journ al h omepage: www .elsevier.co m/lo cate/diabres
Monitoring young people with type 1 diabetes for diabetic nephropathy: Potential errors of annual ACR testing Jason Oke a,*, Andrew Farmer a, Andrew Neil a, R Neil Dalton b, David Dunger b, Richard Stevens a a
Department of Primary Health Care Sciences, University of Oxford and School of Primary Care Research, National Institute for Healthcare Research, United Kingdom b Department of Paediatrics, Institute of Metabolic Science, University of Cambridge, United Kingdom
article info
abstract
Article history:
Aim: Type 1 diabetes guidelines recommend annual monitoring of albumin-creatinine ratio
Received 5 September 2012
(ACR) to detect nephropathy. Annual monitoring for conditions such as dyslipidemia leads
Received in revised form
to high rates of false-positive diagnoses. We estimated rates of false-positive and false-
5 December 2012
negative diagnoses under annual, biennial and triennial monitoring.
Accepted 13 December 2012
Methods: Using Oxford Regional Prospective Study (ORPS) data we modelled ACR over time.
Published on line 9 January 2013
Using simulation we estimated numbers of positive and negative diagnoses and the proportion that are false, over 6 years of monitoring, when assessment intervals are 1, 2
Keywords:
or 3 years.
Monitoring
Results: Average increase per year (95% C.I.) in ACR was 3,5% (2,0–5,0%) for males and 4,8% (3,2–
Diabetes
6,5%) for females. By 6 years, annual monitoring would lead to 56 (49–63) false-positive
Micro-albuminuria
diagnoses for every 100 positive diagnoses of micro-albuminuria, biennial to 49 (42–57)
Nephropathy
false-positives and triennial to 46 (39–53). For every 100 negative diagnoses, annual monitoring would lead to 1,2 (0,8–1,5) false-negatives, biennial to 2,3 (1,7–3,0) and triennial to 3,0 (2,2–3,8). Conclusion: Less frequent monitoring would result in fewer false-positive diagnoses, but increased false negatives, or missed diagnoses. The clinical implications of these scenarios need further investigation through cost-benefit analysis. # 2012 Elsevier Ireland Ltd. All rights reserved.
1.
Introduction
The late stages of diabetic nephropathy, with end-stage renal failure requiring transplant or dialysis, are preceded by many years of asymptomatic deterioration in kidney function, which can be detected by urine albumin testing [1]. In people with diabetes who have elevated urinary albumin levels the progression to more severe nephropathy can be slowed or reduced with angiotensin-converting enzyme (ACE)-inhibitors
or angiotensin receptor blockers (ARBs) [2]. Guidelines from the International Diabetes Federation (IDF) [3], the American Diabetes Association (ADA) [4] and the National Institute for Clinical Excellence (NICE) [5] recommend annual measurement of urine albumin levels in people with type 1 or type 2 diabetes. Urine albumin levels are measured as albumin/creatinine ratio (ACR) in a first-pass urine specimen. Under IDF and NICE guidelines, If ACR is >3,5 mg/mmol for women and 2,5 mg/ mmol for men, then urine albumin levels are considered
* Corresponding author at: Department of Primary Health Care Sciences, New Radcliffe House, Radcliffe Observatory Quarter, Woodstock Road, Oxford OX2 6GG, United Kingdom. Tel.: +44 1865 289297. E-mail address: [email protected] (J. Oke). 0168-8227/$ – see front matter # 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.diabres.2012.12.010
308
diabetes research and clinical practice 99 (2013) 307–314
abnormal. Under the ADA guidelines the threshold for both men and women is 3,4 mg/mmol. NICE, ADA and IDF guidelines recommend that abnormal levels be confirmed with at least one further measurement, i.e. persistent micro-albuminuria [4,5], because ACR has high within-person variability from day-to-day, even compared to other biochemical measurements [4–6]. Annual testing of raised cholesterol has been shown to lead more false positive tests than true positive tests [7,8]. False positive tests occur because biological measures such as cholesterol and urine albumin are naturally variable, and annual testing creates repeated opportunities for false positive tests to occur. In type 1 diabetes, false positive urine albumin tests may lead to premature or unnecessary treatment with ACE inhibitors in people with normal urine albumin levels. In this report we extend the methods of previous studies [7,8] to evaluate annual urine albumin testing in type 1 diabetes while allowing for the recommendation that any single positive test requires verification by at least one subsequent test.
model allows for the possibility that while nephropathy progresses at an average rate in the population, in some individuals it progresses much more rapidly while in others it does not progress at all or may even regress. Model parameters were estimated using WinBUGS [12]. To estimate the rate of true and false positive diagnoses over time in a population similar to the ORPS cohort, we simulated true and observed ACR conditional on the model parameters estimated as described above. In each case we assumed that the monitoring would follow the IDF guidelines for monitoring ACR in people with type 1 diabetes. We use the terms test and measurement to refer to single ACR results and assessment to refer to the collection of tests taken in a year and diagnosis as the overall conclusion based on the tests. The simulation for a single patient proceeds as follows. A baseline log ACR and progression rate parameter are drawn from a multivariate normal distribution with means dependent on their gender and age at diagnosis. These parameters are used to calculate their true log ACR for each year following their diagnosis of diabetes by:
2.
True logðACRÞt ¼ baseline log ACR
Methods
þ annual progression rate A comprehensive description of the population and the research design is detailed elsewhere [9]. Briefly, between 1986 and 1996 young people living within the Oxfordshire Health Authority who were diagnosed with type 1 diabetes before the age of 16 were invited to participate in a longitudinal study with annual assessment of clinical and biochemical measures including three consecutive early morning urine specimens. Urine albumin and creatinine assays were carried out in a central laboratory and the urinary albumin/creatinine ratio (ACR) was calculated for each urine sample. Albumin was measured by an enzyme-linked immunosorbent assay, and creatinine using the modified Jaffe’s method [10]. Annual assessments consisting of three consecutive tests continued for up to 20 years. Written consent was obtained from parents and children were asked to give assent before commencement of the study. Ethical approval was obtained from the district ethics committees in the region. We used statistical modelling to estimate the proportion of positive diagnoses that are false positives, based on the variability of the measure (and similarly for false negative diagnoses). The modelling methods have been established for similar research questions in cholesterol [7], blood pressure [8], HbA1c [11]. We used a longitudinal hierarchical linear model for log ACR with individual intercepts (the initial log ACR value) gradients (change in log ACR per year of age) and within-person variation. The model adjusts the mean baseline value of log ACR by gender and age at diagnosis. The average rate of change in log ACR was adjusted for gender only. We assumed normal distributions for the variation in gradient and intercept between individuals but used a gamma distribution to model the within-person precisions (1/variance) to take into account the extreme ACR measurements that remain even after log transformation. Because we are using log-transformed ACR, estimates for the baseline back-transformed values are geometric means, estimates of progression of ACR represent multiplicative effects and standard deviations are geometric standard deviations. The log-linear, random effects
duration of diabetes ðin yearsÞ
(1)
The IDF guidance is to screen annually for micro-albuminuria from the age of 11 and 2 years duration of diabetes, therefore in the simulation a patient whose age at diagnosis of diabetes is less than 11 years commences monitoring and contributes to the simulation only when they reach age eleven, whereas a patient who is diagnosed older than 11 years starts monitoring 2 years later. Therefore at the beginning of the monitoring scheme patients are ages eleven or older and at least 2 years duration in proportions in line with the ORPS cohort. To simulate a monitoring assessment, up to three observed ACR measurements are derived by adding simulated withinperson variation to the true ACR calculated using Eq. (1). Observed ACRt ¼ True logðACRÞt þ within person variation
(2)
Threshold values of ACR of 2,5 mg/mmol for males and 3,5 mg/ mmol for females were used in the primary analysis and diagnoses at each assessment were determined by (i) if the first urine ACR measurement in a given year is below the threshold, the patient is diagnosed negative and is not measured again that year (ii) if the first test is above threshold, a second urine sample is taken on a separate occasion (iii) if the second urine ACR measurement is also above threshold, the patient is considered to have a positive diagnosis and treatment is indicated (iv) if the second urine ACR measurement is below threshold, a third sample is taken on a third occasion and treatment decision based on whether the third measurement confirms the first (above threshold) or second (below threshold) see Fig. 1. If a patient has a positive diagnosis, they are no longer monitored, but remain accounted for. This allows those whose positive diagnosis is premature to progress from a false positive state to a true positive state. If the diagnosis is negative, the patient continues to be monitored. This was performed 100000 times and the false positive and
309
diabetes research and clinical practice 99 (2013) 307–314
[(Fig._1)TD$IG]
Fig. 1 – Tree diagram of monitoring assessment of ACR for detection of micro-albuminuria. Squares represent terminal actions and circles are ‘‘chance’’ nodes.
false negative rates calculated. The entire simulation was performed three times, with monitoring taking place annually, every 2 years (biennial) and every 3 years (triennial). For uncertainty estimates we further repeated the simulations changing the model parameters to plausible values given the data (posterior distribution). We present the results as means of the simulations and 95% credible intervals calculated by the percentile method. As a sensitivity analysis, we used the American Diabetes Association and NICE guidelines [4,5] for monitoring for microalbuminuria. The ADA and NICE guidelines suggest the same method for diagnoses as the IDF, but differ from the IDF guidelines in when screening should commence. The ADA guidelines are that screening should start 5 years following diagnosis irrespective of age whereas the NICE guidelines state screening should begin at ages 12 irrespective of the duration of diabetes. The ADA guidelines use a common threshold of 3,4 mmol/mol for both women and men. See Appendix 1 (females) and Appendix 2 (males) for results under the ADA guidelines, Appendix 3 (females) and Appendix 4 (males) for results using NICE guidelines for monitoring.
The within-person coefficient of variation (CV) of ACR was estimated using the root mean square method; calculating the mean and within-person variation at each visit when all three measurements of ACR were available and excluded visits if any of the three measurements were missing. For internal validation, we calculated the expected albumin creatinine ratio for each original data point based on the model parameters, patients age, gender and duration of diabetes. We then calculated the number of diagnoses that would be deemed normo-albuminuric, micro-albuminuric or macro-albuminuric under the IDF guideline thresholds. We compared the results of the expected diagnoses to the original data stratified by duration of diabetes.
3.
Results
We used data from 535 patients from a possible 542, excluding seven patients because their duration of diabetes was not recorded. Characteristics of the cohort are presented in Table 1. Subjects were followed up for a median (IQR) 9,0
Table 1 – Characteristics of the modelling cohort. Figures are median (q1, q3) or n (%). eGFR – estimated glomerular filtration rate is calculated from first available measure. Micro-albuminuria is defined by two positives test (threshold for males = 2,5 mmol/mol and 3,5 mmol/mol for women) from three. ACR is the mean of up to three consecutive measurements. Variable
Category
Gender
Male Female
Age at diagnosis (years) Duration of diabetes (in years) HbA1c (mmol/mol) Body mass index (kg/m^2) Systolic blood pressure (mm/Hg) Diastolic blood pressure (mm/Hg) eGFR (ml/min) ACR (mmol/mol) Micro-albuminuria
Yes No
n 296 246 535 535 520 535 537 537 320 542 30 512
Value 54.6% 45.4% 9,6 (5,6–12.1) 1,05 (0,87–1,66) 83 (68–99) 18 (17–20) 100 (92–109) 65 (58–72) 159 (134–182) 0,95 (0,61–1,46) 5,5% 94.5%
310
diabetes research and clinical practice 99 (2013) 307–314
(5,3, 12.3) years, giving a total of 4098 person-years of annual assessments. 179 of the patients were diagnosed with type 1 diabetes between the period of 1985–1990 and 356 were diagnosed between 1990 and 1995. Our model estimates that the geometric mean ACR at baseline (diagnosis of diabetes) is 0,80 (95% C.I. 0,73–0,87) for males and 0,86 (95% C.I. 0,78–0,95) for females. The geometric mean increase in ACR per year for males is estimated to be 3,5% (95% C.I. 2,0–5,0%) and 4,8% (95% C.I. 3,2–6,5%) for females. The between person variation in progression of log ACR is estimated as 0,11 (95% C.I. 0,099– 0,117) for both males and females, this implies that 95% of the male population will progress within a range of a 16% decrease per year to a 28% increase. For females, the variation in progression of ACR ranges from a 15% reduction to a 29% increase per year. The mean within-person standard deviation in log ACR is estimated as 0,79% (95% C.I. 0,73–0,86). The estimated within-person CV of ACR is 48% based on data from 531 patients on a total of 3784 annual visits. Table 2 shows the estimated cumulative numbers of positive and negative diagnoses, percentages of false positive and false negative diagnoses per 1000 females for three monitoring schemes, annual monitoring, biennial and triennial monitoring. In the first year of monitoring we estimate that there would be 63 positive diagnoses and 937 negative diagnoses per 1000 patient (top half of table, top row). Of these 63, we estimate over half (58%, bottom half of table, top row) would be false positive, representing people whose underlying ACR is below the threshold but who have produced two positive diagnoses due to ‘‘chance’’. Of 937 negative diagnoses, we estimate that 1,9% of these would be false negative, those whose underlying ACR is above the threshold but was not detected by the assessment. The numbers for all three schemes are identical in the first year of monitoring. By the sixth year of monitoring, the number of positive diagnoses
(95% C.I) for annual, biennial and triennial schemes are 260 (223–297), 208 (174–242) and 185 (154–217) per 1000 respectively (top half of table, bottom row). The percentage of diagnoses (95% C.I) that are false are 56 (49–63)%, 49 (42–57)% and 46 (39– 53)% (bottom half of table, bottom row). The percentage of negative diagnoses (95% C.I) that remain false in the sixth year are 1,2 (0,8–1,5)%, 2,3 (1,7–3,0) and 3,0 (2,2–3,8)% (bottom half of table, bottom row). Table 3 shows the results of a simulation for males, which differs from the female simulation in that is uses a lower threshold (2,5 mmol/mol) and different estimates of baseline log ACR and progression. Fig. 2 shows the results of the internal validation. In the range at which our analysis is focussed on (approximately 5– 12 years) the expected numbers in each category is comparable to the observed data, in the later stages; at 12 years duration onwards our model over-predicts the number of diagnoses that would be classed as micro-albuminuric at the expense of patients classed as normo-albuminuric.
4.
Discussion
Annual ACR testing, according to IDF guidelines results in a high rate of false positive diagnoses for abnormal levels of urinary albumin. We estimate that approximately one of every two persons positively diagnosed with micro-albuminuria actually has an underlying level that is below current thresholds for diagnosis of micro-albuminuria, and this pattern persists to the end of the six-year period that we considered. Testing urine albumin levels on a triennial basis would reduce the expected number of false positive diagnoses substantially (Tables 2 and 3). A caveat to this is that triennial testing would increase the expected number of false negative diagnoses, but in every analysis we found that false positive
Table 2 – Modelled positive and negative diagnoses per 1000 females with type 1 diabetes (top section of table) and percentages (95% credible intervals) (lower section) of false positive and false negative diagnoses over a 6 year period for annual, biennial and triennial monitoring under IDF guidelines. Cumulative number of positive diagnoses per 1000 (95% confidence interval) Years since first test
Every year
0 1 2 3 4 5 6
63 106 140 172 202 231 260
(50–79) (86–127) (116–166) (144–200) (170–233) (197–266) (223–297)
Every 2 years
Every 3 years
63 (50–79)
63 (50–79)
111 (91–134) 119 (99–142) 158 (131–187) 208 (174–242)
185 (154–217)
Percentage (95% C.I.) of positive diagnoses that are false (95% confidence interval) Years since first test 0 1 2 3 4 5 6
Every year 58 63 64 63 61 59 56
(52–65) (57–69) (58–70) (57–70) (55–68) (52–66) (49–63)
Every 2 years
Every 3 years
58 (52–65)
58 (52–65)
60 (53–66) 56 (49–63) 56 (49–63) 49 (42–57)
46 (39–53)
Cumulative number of negative diagnoses per 1000 (95% confidence interval) Every year 937 894 860 828 798 769 740
(921–950) (873–914) (834–884) (800–856) (767–830) (734–803) (703–777)
Every 2 years
Every 3 years
937 (921–950)
937 (921–950)
889 (866–909) 881 (858–901) 842 (813–869) 792 (758–826)
815 (783–846)
Percentage (95% C.I.) of negative diagnoses that are false (95% confidence interval) Every year 1,9 1,2 0,9 0,9 0,9 1,0 1,2
(1,4–2,6) (0,9–1,5) (0,7–1,2) (0,6–1,1) (0,7–1,2) (0,7–1,4) (0,8–1,5)
Every 2 years
Every 3 years
1,9 (1,4–2,6)
1,9 (1,4–2,6)
1,5 (1,1–2) 2,0 (1,5–2,6) 1,8 (1,3–2,3) 2,3 (1,7–3,0)
3,0 (2,2–3,8)
diabetes research and clinical practice 99 (2013) 307–314
311
Table 3 – Modelled positive and negative diagnoses per 1000 males with type 1 diabetes (top section of table) and percentages (95% credible intervals) (lower section) of false positive and false negative diagnoses over a 6 year period for annual, biennial and triennial monitoring under IDF guidelines. Cumulative number of positive diagnoses per 1000 (95% confidence interval) Years since first test
Every year
0 1 2 3 4 5 6
79 131 175 214 252 290 326
(67–94) (113–153) (152–202) (188–246) (223–287) (256–327) (290–365)
Every 2 years
Every 3 years
79 (67–94)
79 (67–94)
140 (121–163) 153 (132–178) 202 (174–230) 267 (233–299)
240 (210–272)
Percentage (95% C.I.) of positive diagnoses that are false (95% confidence interval) Years since first test 0 1 2 3 4 5 6
Every year 52 57 57 56 54 51 48
(46–59) (51–62) (52–63) (51–62) (49–60) (46–57) (43–54)
Every 2 years
Every 3 years
52 (46–59)
52 (46–59)
53 (47–59) 48 (43–54) 48 (43–54) 42 (37–48)
tests (patients with normal excretion rates misdiagnosed with micro-albuminuria) were many times more prevalent than false negative tests (missed diagnoses of micro-albuminuria). These high rates of false positive tests arise because urine albumin tests are liable to significant variation and extreme observations within the course of the day. The within-person coefficient of variation of ACR in this cohort is 48%. Analytical coefficient of variation of ACR from thawed urine samples has been reported as 6% [13]. Using a standard method [14] we estimate that the biological CV to be 47.6% accounting for 98% of the total variation. The high variability has been previously attributed to several factors including water intake, rate of diuresis, exercise and diet Although there have been reported associations between urinary tract infection and elevated urinary albumin tests, a systematic review recently concluded that there is no evidence to confirm that asymptomatic urinary tract infection causes increases in urinary albumin levels [15]. A significant degree of this biological variation could be explained by current measurements of glycaemic control both within and between patients and incorporated into our model. Although, this would be provide valuable information for setting individualized monitoring intervals we set out to produce unconditional population level estimates. Varying urinary dilution is addressed by the use of albumin creatinine ratio, rather than urine albumin alone, and guidelines [3–5] attempt to address the remaining high variability of ACR by recommending verification of a positive test with further tests prior to initiation of treatment. Both these procedures have been incorporated into our analyses and the rate of false tests, especially false positive tests, remains high. A limitation is the absence of a ‘gold standard’ for diabetic nephropathy: that is, the actual (numbers of) patients who falsely test positive and negative with ACR testing is not known in this dataset. We have overcome this with a statistical modelling approach based on the straightforward,
39 (34–44)
Cumulative number of negative diagnoses per 1000 (95% confidence interval) Every year 921 869 825 786 748 710 674
(906–933) (847–887) (798–848) (754–812) (713–777) (673–744) (635–710)
Every 2 years
Every 3 years
921 (906–933)
921 (906–933)
860 (837–879) 847 (822–868) 798 (770–826) 733 (701–767)
760 (728–790)
Percentage (95% C.I.) of negative diagnoses that are false (95% confidence interval) Every year 2,7 1,7 1,4 1,4 1,5 1,6 1,8
(2,1–3,4) (1,4–2,1) (1,1–1,8) (1,1–1,7) (1,1–1,9) (1,2–2,0) (1,3–2,2)
Every 2 years
Every 3 years
2,7 (2,1–3,4)
2,7 (2,1–3,4)
2,3 (1,8–2,8) 3 (2,4–3,6) 2,7 (2,2–3,3) 3,4 (2,7–4,1)
4,4 (3,6–5,3)
intuitive principle that the proportion of false positive (or negative) tests can be inferred from the variability within the test itself. Although computationally intensive, our methods essentially do little more than extend this principle to allow for the distribution of ACR values in the population, the trends over time, and the recommended use of multiple tests to confirm positive diagnoses of micro-albuminuria. This approach has well-studied statistical properties [11] and has been found useful in the study of similar monitoring tests such as cholesterol, blood pressure, HbA1c and CD4 count [7,8,16,17]. Internal validation diagnostics suggest that the model is a good fit for the time range of the data in which we have calculated the rates of true and false positive diagnoses tests, and discrepancies in the latter stages of duration would make our estimates of the false positive rate conservative (lower). Our analysis is based on a single cohort, predominately not treated with ACEi or ARB’s collected in the UK around 20 years ago. This may limit the generalizability of our analysis to a modern cohort, and in particular to those treated with anti-hypertensives. However, a recent report from a cohort of young people with diabetes in New England remarked explicitly on the similarity of their urine albumin data to the cohort we have used [18] This phenomenon of high numbers of false positive tests occurring with more frequent testing has been observed in other health indicators, as reported by [7] with annual cholesterol testing and HbA1c [16]. Guidelines of the International Diabetes Federation (IDF) [3], American Diabetes Association (ADA) [4] and National Institute for Clinical Excellence (NICE)[5] all agree on annual monitoring for micro-albuminuria but differ on when it should begin; from 5 years duration (ADA), age eleven and 2 years duration (IDF) and age 12 in the NICE guidelines. However, while the IDF and ADA guidelines recommend ACEi for young people with persistent elevated ACR [3,4], no such recommendation exists
312
diabetes research and clinical practice 99 (2013) 307–314
[(Fig._2)TD$IG]
80 60 0
20
40
Prevalence %
60 40 0
20
Prevalence %
80
100
Microalbuminuria
100
Normoalbuminuria
5
10
15
20
Duration of diabetes
5
10
15
20
Duration of diabetes
60 40 0
20
Prevalence %
80
100
Macroalbuminuria
5
10
15
20
Duration of diabetes
Fig. 2 – Internal validation plot. Circles are observed proportions of tests from the ORPS cohort stratified by duration of diabetes that would have been diagnosed normo-albuminuria, micro-albuminuria or macro-albuminuria under current NICE guidelines. Whiskers represent 95% confidence intervals for the proportions using a normal approximation. Dashed line is the expected proportion of diagnoses for the OPRS cohort when observed albumin creatinine ratios are replaced with simulated ACR’s using the model parameter estimates.
in the NICE guidelines [5] for young people with type 1 diabetes and make this recommendation only for adults. The potential benefit of this treatment for adolescents at high risk is the subject of an on-going randomized trial [19]. Regardless of the outcome of this trial, our results indicate that if detection of persistent micro-albuminuria in an annual screening programme were used as an indication for treatment, much of this would be over-treatment due to the high rate of false positives. If ACEi does prove beneficial in a young population then a better method of screening is needed to better target those who will benefit. ACE inhibitor treatment in patients defined, as having normo-albuminuria under current thresholds does not appear to reduce albumin excretion. In a recent systematic review of ACE inhibitor treatment for patients with type 1 diabetes, the average urinary albumin excretion was around 60% lower at follow up for treated micro-albuminuric patients but for patients with normal albumin excretion the reduction was a non-significant 6% [20]. Increasing the interval between assessments from 1 year to 3 years reduces the number of false positive diagnoses but increases the number of false negative diagnoses. Patients who receive a false negative test
result fail to receive the appropriate management until they eventually receive a positive test result. Whether this is a more acceptable option depends on balancing the harm of treating people with normal levels of urinary albumin, which occurs more frequently with annual testing, with the harm in failing to treat people with raised urinary albumin levels; a consequence of extending the interval. The long-term consequences of remaining undiagnosed for prolonged periods is not known but could lead to the development of more severe kidney disease and then end-stage renal disease meaning renal dialysis or transplant. We have assumed that detection follows guidelines exactly, including the ‘best two of three’ algorithm for persistent micro-albuminuria with strict adherence to the thresholds. It has been suggested that tracking the rate of change of albumin excretion might be an alternative approach to the detection of abnormal kidney function in this age group [21]. This is supported by the observation that, in many young people, apparently transient micro-albuminuria often recurs at a later age and intermittent micro-albuminuria is predictive of progression to macro-albuminuria [22]. Further work could investigate whether tracking ACR over a period of time would
diabetes research and clinical practice 99 (2013) 307–314
lead to more reliable detection than periodic monitoring currently recommended by guidelines. However such an investigation would need to consider multiple permutations of monitoring intervals, thresholds and time spans and is beyond the scope of the present paper. A further alternative option for monitoring kidney function with less variability could be to use estimated glomerular filtration rate (eGFR), from serum creatinine levels. The usefulness of estimated GFR alone or as a screening test for kidney disease is limited as many patients with diabetes and chronic kidney disease may have elevated or high-normal, rather than reduced GFRs, particularly in the early years after diagnosis [23]. Urinary albumin is therefore considered a better marker of early kidney damage, while eGFR on its own is suitable for detecting chronic kidney disease at stage 3 or worse [23,24]. In conclusion, less frequent monitoring would result in fewer false positive results, at the expense of an increase in the proportion of false negatives, or missed diagnoses. While false negative tests imply missed opportunities for changes in management, high rates of false positives are of concern because they create unnecessary anxiety and potentially lead to treatment of ACE inhibitors in people with little to gain. The clinical and health consequences of these now need further investigation: for example, by incorporating our results into a full cost-benefit analysis. Better targeting of patients to be monitored by identifying patients most at risk, lowering the cut-off threshold, or repeat testing for all patients including those who test negative, may remove some of the uncertainty in the current monitoring system.
Disclaimer This report is the result of an independent research commissioned by the National Institute for Health Research (NIHR). The views expressed in this publication are those of the authors and not necessarily those of the NHS, the National Institute for Health Research, or the Department of Health.
Conflict of interest The authors declare that there are no conflicts of interest.
Acknowledgments The work in this paper is part funded by NHS Diabetes and the Health Technology Assessment Programme. Andrew Farmer is funded by the National Institute of Health Oxford Biomedical Research Centre. Andrew Neil is a NIHR Senior Investigator. David Dunger is supported by the National Institute for Health Research (NIHR) Cambridge Comprehensive Biomedical Research Centre. We would like to acknowledge the use of the Oxford Supercomputing Centre (OSC) in carrying out this work and would like to thank Jennifer Hirst and Tom Lung for their help with this study.
313
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.diabres.2012.12.010.
references
[1] Mogensen C, Chachati A, Christensen C, Close C, Deckert T, Hommel E, et al. Microalbuminuria: an early marker of renal involvement in diabetes. Uremia Invest 1985;9(2):85–95. [2] Strippoli G, Bonifati C, Craig M, Navaneethan S, Craig J. Angiotensin converting enzyme inhibitors and angiotensin II receptor antagonists for preventing the progression of diabetic kidney disease. Cochrane Database Syst Rev 2006;4:pCD006257. [3] International Diabetes Federation. Global IDF/ISPAD guidline for diabetes in childhood; 2011. [4] American Diabetes Association. Standards of medical care in diabetes – 009. Diabetes Care 2009;32(Suppl. 1):S13–61. [5] National Institute for Clinical Excellence (NICE). Clinical guideline 15: type 1 diabetes: diagnosis management of type 1 diabetes in children young people adults; 2011. [6] Miller W, Bruns D, Hortin G, Sandberg S, Aakre K, McQueen M, et al. Current issues in measurement and reporting of urinary albumin excretion. Clin Chem 2009;55(1):24–38. [7] Glasziou P, Irwig L, Heritier S, Simes R, Tonkin A, LIPID Study Investigators. Monitoring cholesterol levels: measurement error or true change? Ann Intern Med 2008;148(9):656–61. [8] Keenan K, Hayen A, Neal B, Irwig L. Long term monitoring in patients receiving treatment to lower blood pressure: analysis of data from placebo controlled randomised controlled trial. BMJ 2009;338:pb1492. [9] Schultz C, Konopelska-Bahu T, Dalton R, Carroll T, Stratton I, Gale E, et al. Microalbuminuria prevalence varies with age, sex, and puberty in children with type 1 diabetes followed from diagnosis in a longitudinal study Oxford Regional Prospective Study Group. Diabetes Care 1999;22(3):495–502. [10] Husdan H, Rapoport A. Estimation of creatinine by the jaffe reaction. Clin Chem 1968;14(3):222–38. [11] Stevens RJ, Oke J, Perera R. Statistical models for the control phase of clinical monitoring. Stat Methods Med Res 2010;19(4):394–414. [12] Lunn D, Thomas A, Best N, Spiegelhalter D. WinBUGS – a bayesian modelling framework: concepts, structure, and extensibility. Stat Comput 2000;10(4):325–37. [13] Mosca A, Paleari R, Ceriotti F, Lapolla A, Fedele D, Lercanidipine Study Group. Biological variability of albumin excretion rate and albumin-to-creatinine ratio in hypertensive type 2 diabetic patients. Clin Chem Lab Med 2003;41(9):1229–33. [14] Franzini C. Relevance of analytical and biological variations to quality and interpretation of test results: examples of application to haematology. Ann Ist Super Sanita 1995;31(1):9–13. [15] Carter J, Tomson C, Stevens P, Lamb E. Does urinary tract infection cause proteinuria or microalbuminuria?. A systematic review. Nephrol Dial Transplant 2006;21(11):3031–7. [16] Oke J, Stevens R, Gaitskell K, Farmer A. Establishing an evidence base for frequency of monitoring glycated haemoglobin levels in patients with type 2 diabetes:
314
[17]
[18]
[19]
[20]
diabetes research and clinical practice 99 (2013) 307–314
projections of effectiveness from a regression model. Diabetic Med 2012;29(2):266–71. Buclin T, Telenti A, Perera R, Csajka C, Furrer H, Aronson J, et al. Development and validation of decision rules to guide frequency of monitoring CD4 cell count in HIV-1 infection before starting antiretroviral therapy. PLoS One 2011;6(4):pe18578. Alleyn C, Volkening L, Wolfson J, Rodriguez-Ventura A, Wood J, Laffel L. Occurrence of microalbuminuria in young people with Type 1 diabetes: importance of age and diabetes duration. Diabetic Med 2010;27(5):532–7. The adolescent type 1 diabetes cardio-renal intervention trial research group. Adolescent type 1 Diabetes Cardiorenal Intervention Trial (AdDIT). BMC Pediatr 2009;9:p79. Hirst J, Taylor K, Stevens R, Blacklock C, Roberts N, Pugh C, et al. The impact of renin-angiotensin-aldosterone system inhibitors on Type 1 and Type 2 diabetic patients with and without early diabetic nephropathy. Kidney Int 2012;81(7):674–83.
[21] Dunger D, Schwarze C, Cooper J, Widmer B, Neil A, Shield J, et al. Can we identify adolescents at high risk for nephropathy before the development of microalbuminuria? Diabetic Med 2007;24(2):131–6. [22] Amin R, Widmer B, Prevost T, Schwarze P, Cooper J, Edge J, et al. Risk of microalbuminuria and progression to macroalbuminuria in a cohort with childhood onset type 1 diabetes: prospective observational study. BMJ (Clinical research ed) 2008;336(7646):697–701. [23] KDOQI. KDOQI Clinical practice guidelines and clinical practice recommendations for diabetes and chronic kidney disease. Am J Kidney Dis 2007; 49(Suppl. 2): p. S12-154. [24] Marcovecchio M, Dalton R, Turner C, Prevost T, Widmer B, Amin R, et al. Symmetric dimethylarginine, an endogenous marker of glomerular filtration rate, and the risk for microalbuminuria in young people with type 1 diabetes. Arch Dis Child 2010;95(2): 119–24.
Contents available at Sciverse ScienceDirect
Diabetes Research and Clinical Practice journ al h omepage: www .elsevier.co m/lo cate/diabres
Monitoring young people with type 1 diabetes for diabetic nephropathy: Potential errors of annual ACR testing Jason Oke a,*, Andrew Farmer a, Andrew Neil a, R Neil Dalton b, David Dunger b, Richard Stevens a a
Department of Primary Health Care Sciences, University of Oxford and School of Primary Care Research, National Institute for Healthcare Research, United Kingdom b Department of Paediatrics, Institute of Metabolic Science, University of Cambridge, United Kingdom
article info
abstract
Article history:
Aim: Type 1 diabetes guidelines recommend annual monitoring of albumin-creatinine ratio
Received 5 September 2012
(ACR) to detect nephropathy. Annual monitoring for conditions such as dyslipidemia leads
Received in revised form
to high rates of false-positive diagnoses. We estimated rates of false-positive and false-
5 December 2012
negative diagnoses under annual, biennial and triennial monitoring.
Accepted 13 December 2012
Methods: Using Oxford Regional Prospective Study (ORPS) data we modelled ACR over time.
Published on line 9 January 2013
Using simulation we estimated numbers of positive and negative diagnoses and the proportion that are false, over 6 years of monitoring, when assessment intervals are 1, 2
Keywords:
or 3 years.
Monitoring
Results: Average increase per year (95% C.I.) in ACR was 3,5% (2,0–5,0%) for males and 4,8% (3,2–
Diabetes
6,5%) for females. By 6 years, annual monitoring would lead to 56 (49–63) false-positive
Micro-albuminuria
diagnoses for every 100 positive diagnoses of micro-albuminuria, biennial to 49 (42–57)
Nephropathy
false-positives and triennial to 46 (39–53). For every 100 negative diagnoses, annual monitoring would lead to 1,2 (0,8–1,5) false-negatives, biennial to 2,3 (1,7–3,0) and triennial to 3,0 (2,2–3,8). Conclusion: Less frequent monitoring would result in fewer false-positive diagnoses, but increased false negatives, or missed diagnoses. The clinical implications of these scenarios need further investigation through cost-benefit analysis. # 2012 Elsevier Ireland Ltd. All rights reserved.
1.
Introduction
The late stages of diabetic nephropathy, with end-stage renal failure requiring transplant or dialysis, are preceded by many years of asymptomatic deterioration in kidney function, which can be detected by urine albumin testing [1]. In people with diabetes who have elevated urinary albumin levels the progression to more severe nephropathy can be slowed or reduced with angiotensin-converting enzyme (ACE)-inhibitors
or angiotensin receptor blockers (ARBs) [2]. Guidelines from the International Diabetes Federation (IDF) [3], the American Diabetes Association (ADA) [4] and the National Institute for Clinical Excellence (NICE) [5] recommend annual measurement of urine albumin levels in people with type 1 or type 2 diabetes. Urine albumin levels are measured as albumin/creatinine ratio (ACR) in a first-pass urine specimen. Under IDF and NICE guidelines, If ACR is >3,5 mg/mmol for women and 2,5 mg/ mmol for men, then urine albumin levels are considered
* Corresponding author at: Department of Primary Health Care Sciences, New Radcliffe House, Radcliffe Observatory Quarter, Woodstock Road, Oxford OX2 6GG, United Kingdom. Tel.: +44 1865 289297. E-mail address: [email protected] (J. Oke). 0168-8227/$ – see front matter # 2012 Elsevier Ireland Ltd. All rights reserved. http://dx.doi.org/10.1016/j.diabres.2012.12.010
308
diabetes research and clinical practice 99 (2013) 307–314
abnormal. Under the ADA guidelines the threshold for both men and women is 3,4 mg/mmol. NICE, ADA and IDF guidelines recommend that abnormal levels be confirmed with at least one further measurement, i.e. persistent micro-albuminuria [4,5], because ACR has high within-person variability from day-to-day, even compared to other biochemical measurements [4–6]. Annual testing of raised cholesterol has been shown to lead more false positive tests than true positive tests [7,8]. False positive tests occur because biological measures such as cholesterol and urine albumin are naturally variable, and annual testing creates repeated opportunities for false positive tests to occur. In type 1 diabetes, false positive urine albumin tests may lead to premature or unnecessary treatment with ACE inhibitors in people with normal urine albumin levels. In this report we extend the methods of previous studies [7,8] to evaluate annual urine albumin testing in type 1 diabetes while allowing for the recommendation that any single positive test requires verification by at least one subsequent test.
model allows for the possibility that while nephropathy progresses at an average rate in the population, in some individuals it progresses much more rapidly while in others it does not progress at all or may even regress. Model parameters were estimated using WinBUGS [12]. To estimate the rate of true and false positive diagnoses over time in a population similar to the ORPS cohort, we simulated true and observed ACR conditional on the model parameters estimated as described above. In each case we assumed that the monitoring would follow the IDF guidelines for monitoring ACR in people with type 1 diabetes. We use the terms test and measurement to refer to single ACR results and assessment to refer to the collection of tests taken in a year and diagnosis as the overall conclusion based on the tests. The simulation for a single patient proceeds as follows. A baseline log ACR and progression rate parameter are drawn from a multivariate normal distribution with means dependent on their gender and age at diagnosis. These parameters are used to calculate their true log ACR for each year following their diagnosis of diabetes by:
2.
True logðACRÞt ¼ baseline log ACR
Methods
þ annual progression rate A comprehensive description of the population and the research design is detailed elsewhere [9]. Briefly, between 1986 and 1996 young people living within the Oxfordshire Health Authority who were diagnosed with type 1 diabetes before the age of 16 were invited to participate in a longitudinal study with annual assessment of clinical and biochemical measures including three consecutive early morning urine specimens. Urine albumin and creatinine assays were carried out in a central laboratory and the urinary albumin/creatinine ratio (ACR) was calculated for each urine sample. Albumin was measured by an enzyme-linked immunosorbent assay, and creatinine using the modified Jaffe’s method [10]. Annual assessments consisting of three consecutive tests continued for up to 20 years. Written consent was obtained from parents and children were asked to give assent before commencement of the study. Ethical approval was obtained from the district ethics committees in the region. We used statistical modelling to estimate the proportion of positive diagnoses that are false positives, based on the variability of the measure (and similarly for false negative diagnoses). The modelling methods have been established for similar research questions in cholesterol [7], blood pressure [8], HbA1c [11]. We used a longitudinal hierarchical linear model for log ACR with individual intercepts (the initial log ACR value) gradients (change in log ACR per year of age) and within-person variation. The model adjusts the mean baseline value of log ACR by gender and age at diagnosis. The average rate of change in log ACR was adjusted for gender only. We assumed normal distributions for the variation in gradient and intercept between individuals but used a gamma distribution to model the within-person precisions (1/variance) to take into account the extreme ACR measurements that remain even after log transformation. Because we are using log-transformed ACR, estimates for the baseline back-transformed values are geometric means, estimates of progression of ACR represent multiplicative effects and standard deviations are geometric standard deviations. The log-linear, random effects
duration of diabetes ðin yearsÞ
(1)
The IDF guidance is to screen annually for micro-albuminuria from the age of 11 and 2 years duration of diabetes, therefore in the simulation a patient whose age at diagnosis of diabetes is less than 11 years commences monitoring and contributes to the simulation only when they reach age eleven, whereas a patient who is diagnosed older than 11 years starts monitoring 2 years later. Therefore at the beginning of the monitoring scheme patients are ages eleven or older and at least 2 years duration in proportions in line with the ORPS cohort. To simulate a monitoring assessment, up to three observed ACR measurements are derived by adding simulated withinperson variation to the true ACR calculated using Eq. (1). Observed ACRt ¼ True logðACRÞt þ within person variation
(2)
Threshold values of ACR of 2,5 mg/mmol for males and 3,5 mg/ mmol for females were used in the primary analysis and diagnoses at each assessment were determined by (i) if the first urine ACR measurement in a given year is below the threshold, the patient is diagnosed negative and is not measured again that year (ii) if the first test is above threshold, a second urine sample is taken on a separate occasion (iii) if the second urine ACR measurement is also above threshold, the patient is considered to have a positive diagnosis and treatment is indicated (iv) if the second urine ACR measurement is below threshold, a third sample is taken on a third occasion and treatment decision based on whether the third measurement confirms the first (above threshold) or second (below threshold) see Fig. 1. If a patient has a positive diagnosis, they are no longer monitored, but remain accounted for. This allows those whose positive diagnosis is premature to progress from a false positive state to a true positive state. If the diagnosis is negative, the patient continues to be monitored. This was performed 100000 times and the false positive and
309
diabetes research and clinical practice 99 (2013) 307–314
[(Fig._1)TD$IG]
Fig. 1 – Tree diagram of monitoring assessment of ACR for detection of micro-albuminuria. Squares represent terminal actions and circles are ‘‘chance’’ nodes.
false negative rates calculated. The entire simulation was performed three times, with monitoring taking place annually, every 2 years (biennial) and every 3 years (triennial). For uncertainty estimates we further repeated the simulations changing the model parameters to plausible values given the data (posterior distribution). We present the results as means of the simulations and 95% credible intervals calculated by the percentile method. As a sensitivity analysis, we used the American Diabetes Association and NICE guidelines [4,5] for monitoring for microalbuminuria. The ADA and NICE guidelines suggest the same method for diagnoses as the IDF, but differ from the IDF guidelines in when screening should commence. The ADA guidelines are that screening should start 5 years following diagnosis irrespective of age whereas the NICE guidelines state screening should begin at ages 12 irrespective of the duration of diabetes. The ADA guidelines use a common threshold of 3,4 mmol/mol for both women and men. See Appendix 1 (females) and Appendix 2 (males) for results under the ADA guidelines, Appendix 3 (females) and Appendix 4 (males) for results using NICE guidelines for monitoring.
The within-person coefficient of variation (CV) of ACR was estimated using the root mean square method; calculating the mean and within-person variation at each visit when all three measurements of ACR were available and excluded visits if any of the three measurements were missing. For internal validation, we calculated the expected albumin creatinine ratio for each original data point based on the model parameters, patients age, gender and duration of diabetes. We then calculated the number of diagnoses that would be deemed normo-albuminuric, micro-albuminuric or macro-albuminuric under the IDF guideline thresholds. We compared the results of the expected diagnoses to the original data stratified by duration of diabetes.
3.
Results
We used data from 535 patients from a possible 542, excluding seven patients because their duration of diabetes was not recorded. Characteristics of the cohort are presented in Table 1. Subjects were followed up for a median (IQR) 9,0
Table 1 – Characteristics of the modelling cohort. Figures are median (q1, q3) or n (%). eGFR – estimated glomerular filtration rate is calculated from first available measure. Micro-albuminuria is defined by two positives test (threshold for males = 2,5 mmol/mol and 3,5 mmol/mol for women) from three. ACR is the mean of up to three consecutive measurements. Variable
Category
Gender
Male Female
Age at diagnosis (years) Duration of diabetes (in years) HbA1c (mmol/mol) Body mass index (kg/m^2) Systolic blood pressure (mm/Hg) Diastolic blood pressure (mm/Hg) eGFR (ml/min) ACR (mmol/mol) Micro-albuminuria
Yes No
n 296 246 535 535 520 535 537 537 320 542 30 512
Value 54.6% 45.4% 9,6 (5,6–12.1) 1,05 (0,87–1,66) 83 (68–99) 18 (17–20) 100 (92–109) 65 (58–72) 159 (134–182) 0,95 (0,61–1,46) 5,5% 94.5%
310
diabetes research and clinical practice 99 (2013) 307–314
(5,3, 12.3) years, giving a total of 4098 person-years of annual assessments. 179 of the patients were diagnosed with type 1 diabetes between the period of 1985–1990 and 356 were diagnosed between 1990 and 1995. Our model estimates that the geometric mean ACR at baseline (diagnosis of diabetes) is 0,80 (95% C.I. 0,73–0,87) for males and 0,86 (95% C.I. 0,78–0,95) for females. The geometric mean increase in ACR per year for males is estimated to be 3,5% (95% C.I. 2,0–5,0%) and 4,8% (95% C.I. 3,2–6,5%) for females. The between person variation in progression of log ACR is estimated as 0,11 (95% C.I. 0,099– 0,117) for both males and females, this implies that 95% of the male population will progress within a range of a 16% decrease per year to a 28% increase. For females, the variation in progression of ACR ranges from a 15% reduction to a 29% increase per year. The mean within-person standard deviation in log ACR is estimated as 0,79% (95% C.I. 0,73–0,86). The estimated within-person CV of ACR is 48% based on data from 531 patients on a total of 3784 annual visits. Table 2 shows the estimated cumulative numbers of positive and negative diagnoses, percentages of false positive and false negative diagnoses per 1000 females for three monitoring schemes, annual monitoring, biennial and triennial monitoring. In the first year of monitoring we estimate that there would be 63 positive diagnoses and 937 negative diagnoses per 1000 patient (top half of table, top row). Of these 63, we estimate over half (58%, bottom half of table, top row) would be false positive, representing people whose underlying ACR is below the threshold but who have produced two positive diagnoses due to ‘‘chance’’. Of 937 negative diagnoses, we estimate that 1,9% of these would be false negative, those whose underlying ACR is above the threshold but was not detected by the assessment. The numbers for all three schemes are identical in the first year of monitoring. By the sixth year of monitoring, the number of positive diagnoses
(95% C.I) for annual, biennial and triennial schemes are 260 (223–297), 208 (174–242) and 185 (154–217) per 1000 respectively (top half of table, bottom row). The percentage of diagnoses (95% C.I) that are false are 56 (49–63)%, 49 (42–57)% and 46 (39– 53)% (bottom half of table, bottom row). The percentage of negative diagnoses (95% C.I) that remain false in the sixth year are 1,2 (0,8–1,5)%, 2,3 (1,7–3,0) and 3,0 (2,2–3,8)% (bottom half of table, bottom row). Table 3 shows the results of a simulation for males, which differs from the female simulation in that is uses a lower threshold (2,5 mmol/mol) and different estimates of baseline log ACR and progression. Fig. 2 shows the results of the internal validation. In the range at which our analysis is focussed on (approximately 5– 12 years) the expected numbers in each category is comparable to the observed data, in the later stages; at 12 years duration onwards our model over-predicts the number of diagnoses that would be classed as micro-albuminuric at the expense of patients classed as normo-albuminuric.
4.
Discussion
Annual ACR testing, according to IDF guidelines results in a high rate of false positive diagnoses for abnormal levels of urinary albumin. We estimate that approximately one of every two persons positively diagnosed with micro-albuminuria actually has an underlying level that is below current thresholds for diagnosis of micro-albuminuria, and this pattern persists to the end of the six-year period that we considered. Testing urine albumin levels on a triennial basis would reduce the expected number of false positive diagnoses substantially (Tables 2 and 3). A caveat to this is that triennial testing would increase the expected number of false negative diagnoses, but in every analysis we found that false positive
Table 2 – Modelled positive and negative diagnoses per 1000 females with type 1 diabetes (top section of table) and percentages (95% credible intervals) (lower section) of false positive and false negative diagnoses over a 6 year period for annual, biennial and triennial monitoring under IDF guidelines. Cumulative number of positive diagnoses per 1000 (95% confidence interval) Years since first test
Every year
0 1 2 3 4 5 6
63 106 140 172 202 231 260
(50–79) (86–127) (116–166) (144–200) (170–233) (197–266) (223–297)
Every 2 years
Every 3 years
63 (50–79)
63 (50–79)
111 (91–134) 119 (99–142) 158 (131–187) 208 (174–242)
185 (154–217)
Percentage (95% C.I.) of positive diagnoses that are false (95% confidence interval) Years since first test 0 1 2 3 4 5 6
Every year 58 63 64 63 61 59 56
(52–65) (57–69) (58–70) (57–70) (55–68) (52–66) (49–63)
Every 2 years
Every 3 years
58 (52–65)
58 (52–65)
60 (53–66) 56 (49–63) 56 (49–63) 49 (42–57)
46 (39–53)
Cumulative number of negative diagnoses per 1000 (95% confidence interval) Every year 937 894 860 828 798 769 740
(921–950) (873–914) (834–884) (800–856) (767–830) (734–803) (703–777)
Every 2 years
Every 3 years
937 (921–950)
937 (921–950)
889 (866–909) 881 (858–901) 842 (813–869) 792 (758–826)
815 (783–846)
Percentage (95% C.I.) of negative diagnoses that are false (95% confidence interval) Every year 1,9 1,2 0,9 0,9 0,9 1,0 1,2
(1,4–2,6) (0,9–1,5) (0,7–1,2) (0,6–1,1) (0,7–1,2) (0,7–1,4) (0,8–1,5)
Every 2 years
Every 3 years
1,9 (1,4–2,6)
1,9 (1,4–2,6)
1,5 (1,1–2) 2,0 (1,5–2,6) 1,8 (1,3–2,3) 2,3 (1,7–3,0)
3,0 (2,2–3,8)
diabetes research and clinical practice 99 (2013) 307–314
311
Table 3 – Modelled positive and negative diagnoses per 1000 males with type 1 diabetes (top section of table) and percentages (95% credible intervals) (lower section) of false positive and false negative diagnoses over a 6 year period for annual, biennial and triennial monitoring under IDF guidelines. Cumulative number of positive diagnoses per 1000 (95% confidence interval) Years since first test
Every year
0 1 2 3 4 5 6
79 131 175 214 252 290 326
(67–94) (113–153) (152–202) (188–246) (223–287) (256–327) (290–365)
Every 2 years
Every 3 years
79 (67–94)
79 (67–94)
140 (121–163) 153 (132–178) 202 (174–230) 267 (233–299)
240 (210–272)
Percentage (95% C.I.) of positive diagnoses that are false (95% confidence interval) Years since first test 0 1 2 3 4 5 6
Every year 52 57 57 56 54 51 48
(46–59) (51–62) (52–63) (51–62) (49–60) (46–57) (43–54)
Every 2 years
Every 3 years
52 (46–59)
52 (46–59)
53 (47–59) 48 (43–54) 48 (43–54) 42 (37–48)
tests (patients with normal excretion rates misdiagnosed with micro-albuminuria) were many times more prevalent than false negative tests (missed diagnoses of micro-albuminuria). These high rates of false positive tests arise because urine albumin tests are liable to significant variation and extreme observations within the course of the day. The within-person coefficient of variation of ACR in this cohort is 48%. Analytical coefficient of variation of ACR from thawed urine samples has been reported as 6% [13]. Using a standard method [14] we estimate that the biological CV to be 47.6% accounting for 98% of the total variation. The high variability has been previously attributed to several factors including water intake, rate of diuresis, exercise and diet Although there have been reported associations between urinary tract infection and elevated urinary albumin tests, a systematic review recently concluded that there is no evidence to confirm that asymptomatic urinary tract infection causes increases in urinary albumin levels [15]. A significant degree of this biological variation could be explained by current measurements of glycaemic control both within and between patients and incorporated into our model. Although, this would be provide valuable information for setting individualized monitoring intervals we set out to produce unconditional population level estimates. Varying urinary dilution is addressed by the use of albumin creatinine ratio, rather than urine albumin alone, and guidelines [3–5] attempt to address the remaining high variability of ACR by recommending verification of a positive test with further tests prior to initiation of treatment. Both these procedures have been incorporated into our analyses and the rate of false tests, especially false positive tests, remains high. A limitation is the absence of a ‘gold standard’ for diabetic nephropathy: that is, the actual (numbers of) patients who falsely test positive and negative with ACR testing is not known in this dataset. We have overcome this with a statistical modelling approach based on the straightforward,
39 (34–44)
Cumulative number of negative diagnoses per 1000 (95% confidence interval) Every year 921 869 825 786 748 710 674
(906–933) (847–887) (798–848) (754–812) (713–777) (673–744) (635–710)
Every 2 years
Every 3 years
921 (906–933)
921 (906–933)
860 (837–879) 847 (822–868) 798 (770–826) 733 (701–767)
760 (728–790)
Percentage (95% C.I.) of negative diagnoses that are false (95% confidence interval) Every year 2,7 1,7 1,4 1,4 1,5 1,6 1,8
(2,1–3,4) (1,4–2,1) (1,1–1,8) (1,1–1,7) (1,1–1,9) (1,2–2,0) (1,3–2,2)
Every 2 years
Every 3 years
2,7 (2,1–3,4)
2,7 (2,1–3,4)
2,3 (1,8–2,8) 3 (2,4–3,6) 2,7 (2,2–3,3) 3,4 (2,7–4,1)
4,4 (3,6–5,3)
intuitive principle that the proportion of false positive (or negative) tests can be inferred from the variability within the test itself. Although computationally intensive, our methods essentially do little more than extend this principle to allow for the distribution of ACR values in the population, the trends over time, and the recommended use of multiple tests to confirm positive diagnoses of micro-albuminuria. This approach has well-studied statistical properties [11] and has been found useful in the study of similar monitoring tests such as cholesterol, blood pressure, HbA1c and CD4 count [7,8,16,17]. Internal validation diagnostics suggest that the model is a good fit for the time range of the data in which we have calculated the rates of true and false positive diagnoses tests, and discrepancies in the latter stages of duration would make our estimates of the false positive rate conservative (lower). Our analysis is based on a single cohort, predominately not treated with ACEi or ARB’s collected in the UK around 20 years ago. This may limit the generalizability of our analysis to a modern cohort, and in particular to those treated with anti-hypertensives. However, a recent report from a cohort of young people with diabetes in New England remarked explicitly on the similarity of their urine albumin data to the cohort we have used [18] This phenomenon of high numbers of false positive tests occurring with more frequent testing has been observed in other health indicators, as reported by [7] with annual cholesterol testing and HbA1c [16]. Guidelines of the International Diabetes Federation (IDF) [3], American Diabetes Association (ADA) [4] and National Institute for Clinical Excellence (NICE)[5] all agree on annual monitoring for micro-albuminuria but differ on when it should begin; from 5 years duration (ADA), age eleven and 2 years duration (IDF) and age 12 in the NICE guidelines. However, while the IDF and ADA guidelines recommend ACEi for young people with persistent elevated ACR [3,4], no such recommendation exists
312
diabetes research and clinical practice 99 (2013) 307–314
[(Fig._2)TD$IG]
80 60 0
20
40
Prevalence %
60 40 0
20
Prevalence %
80
100
Microalbuminuria
100
Normoalbuminuria
5
10
15
20
Duration of diabetes
5
10
15
20
Duration of diabetes
60 40 0
20
Prevalence %
80
100
Macroalbuminuria
5
10
15
20
Duration of diabetes
Fig. 2 – Internal validation plot. Circles are observed proportions of tests from the ORPS cohort stratified by duration of diabetes that would have been diagnosed normo-albuminuria, micro-albuminuria or macro-albuminuria under current NICE guidelines. Whiskers represent 95% confidence intervals for the proportions using a normal approximation. Dashed line is the expected proportion of diagnoses for the OPRS cohort when observed albumin creatinine ratios are replaced with simulated ACR’s using the model parameter estimates.
in the NICE guidelines [5] for young people with type 1 diabetes and make this recommendation only for adults. The potential benefit of this treatment for adolescents at high risk is the subject of an on-going randomized trial [19]. Regardless of the outcome of this trial, our results indicate that if detection of persistent micro-albuminuria in an annual screening programme were used as an indication for treatment, much of this would be over-treatment due to the high rate of false positives. If ACEi does prove beneficial in a young population then a better method of screening is needed to better target those who will benefit. ACE inhibitor treatment in patients defined, as having normo-albuminuria under current thresholds does not appear to reduce albumin excretion. In a recent systematic review of ACE inhibitor treatment for patients with type 1 diabetes, the average urinary albumin excretion was around 60% lower at follow up for treated micro-albuminuric patients but for patients with normal albumin excretion the reduction was a non-significant 6% [20]. Increasing the interval between assessments from 1 year to 3 years reduces the number of false positive diagnoses but increases the number of false negative diagnoses. Patients who receive a false negative test
result fail to receive the appropriate management until they eventually receive a positive test result. Whether this is a more acceptable option depends on balancing the harm of treating people with normal levels of urinary albumin, which occurs more frequently with annual testing, with the harm in failing to treat people with raised urinary albumin levels; a consequence of extending the interval. The long-term consequences of remaining undiagnosed for prolonged periods is not known but could lead to the development of more severe kidney disease and then end-stage renal disease meaning renal dialysis or transplant. We have assumed that detection follows guidelines exactly, including the ‘best two of three’ algorithm for persistent micro-albuminuria with strict adherence to the thresholds. It has been suggested that tracking the rate of change of albumin excretion might be an alternative approach to the detection of abnormal kidney function in this age group [21]. This is supported by the observation that, in many young people, apparently transient micro-albuminuria often recurs at a later age and intermittent micro-albuminuria is predictive of progression to macro-albuminuria [22]. Further work could investigate whether tracking ACR over a period of time would
diabetes research and clinical practice 99 (2013) 307–314
lead to more reliable detection than periodic monitoring currently recommended by guidelines. However such an investigation would need to consider multiple permutations of monitoring intervals, thresholds and time spans and is beyond the scope of the present paper. A further alternative option for monitoring kidney function with less variability could be to use estimated glomerular filtration rate (eGFR), from serum creatinine levels. The usefulness of estimated GFR alone or as a screening test for kidney disease is limited as many patients with diabetes and chronic kidney disease may have elevated or high-normal, rather than reduced GFRs, particularly in the early years after diagnosis [23]. Urinary albumin is therefore considered a better marker of early kidney damage, while eGFR on its own is suitable for detecting chronic kidney disease at stage 3 or worse [23,24]. In conclusion, less frequent monitoring would result in fewer false positive results, at the expense of an increase in the proportion of false negatives, or missed diagnoses. While false negative tests imply missed opportunities for changes in management, high rates of false positives are of concern because they create unnecessary anxiety and potentially lead to treatment of ACE inhibitors in people with little to gain. The clinical and health consequences of these now need further investigation: for example, by incorporating our results into a full cost-benefit analysis. Better targeting of patients to be monitored by identifying patients most at risk, lowering the cut-off threshold, or repeat testing for all patients including those who test negative, may remove some of the uncertainty in the current monitoring system.
Disclaimer This report is the result of an independent research commissioned by the National Institute for Health Research (NIHR). The views expressed in this publication are those of the authors and not necessarily those of the NHS, the National Institute for Health Research, or the Department of Health.
Conflict of interest The authors declare that there are no conflicts of interest.
Acknowledgments The work in this paper is part funded by NHS Diabetes and the Health Technology Assessment Programme. Andrew Farmer is funded by the National Institute of Health Oxford Biomedical Research Centre. Andrew Neil is a NIHR Senior Investigator. David Dunger is supported by the National Institute for Health Research (NIHR) Cambridge Comprehensive Biomedical Research Centre. We would like to acknowledge the use of the Oxford Supercomputing Centre (OSC) in carrying out this work and would like to thank Jennifer Hirst and Tom Lung for their help with this study.
313
Appendix A. Supplementary data Supplementary data associated with this article can be found, in the online version, at http://dx.doi.org/10.1016/ j.diabres.2012.12.010.
references
[1] Mogensen C, Chachati A, Christensen C, Close C, Deckert T, Hommel E, et al. Microalbuminuria: an early marker of renal involvement in diabetes. Uremia Invest 1985;9(2):85–95. [2] Strippoli G, Bonifati C, Craig M, Navaneethan S, Craig J. Angiotensin converting enzyme inhibitors and angiotensin II receptor antagonists for preventing the progression of diabetic kidney disease. Cochrane Database Syst Rev 2006;4:pCD006257. [3] International Diabetes Federation. Global IDF/ISPAD guidline for diabetes in childhood; 2011. [4] American Diabetes Association. Standards of medical care in diabetes – 009. Diabetes Care 2009;32(Suppl. 1):S13–61. [5] National Institute for Clinical Excellence (NICE). Clinical guideline 15: type 1 diabetes: diagnosis management of type 1 diabetes in children young people adults; 2011. [6] Miller W, Bruns D, Hortin G, Sandberg S, Aakre K, McQueen M, et al. Current issues in measurement and reporting of urinary albumin excretion. Clin Chem 2009;55(1):24–38. [7] Glasziou P, Irwig L, Heritier S, Simes R, Tonkin A, LIPID Study Investigators. Monitoring cholesterol levels: measurement error or true change? Ann Intern Med 2008;148(9):656–61. [8] Keenan K, Hayen A, Neal B, Irwig L. Long term monitoring in patients receiving treatment to lower blood pressure: analysis of data from placebo controlled randomised controlled trial. BMJ 2009;338:pb1492. [9] Schultz C, Konopelska-Bahu T, Dalton R, Carroll T, Stratton I, Gale E, et al. Microalbuminuria prevalence varies with age, sex, and puberty in children with type 1 diabetes followed from diagnosis in a longitudinal study Oxford Regional Prospective Study Group. Diabetes Care 1999;22(3):495–502. [10] Husdan H, Rapoport A. Estimation of creatinine by the jaffe reaction. Clin Chem 1968;14(3):222–38. [11] Stevens RJ, Oke J, Perera R. Statistical models for the control phase of clinical monitoring. Stat Methods Med Res 2010;19(4):394–414. [12] Lunn D, Thomas A, Best N, Spiegelhalter D. WinBUGS – a bayesian modelling framework: concepts, structure, and extensibility. Stat Comput 2000;10(4):325–37. [13] Mosca A, Paleari R, Ceriotti F, Lapolla A, Fedele D, Lercanidipine Study Group. Biological variability of albumin excretion rate and albumin-to-creatinine ratio in hypertensive type 2 diabetic patients. Clin Chem Lab Med 2003;41(9):1229–33. [14] Franzini C. Relevance of analytical and biological variations to quality and interpretation of test results: examples of application to haematology. Ann Ist Super Sanita 1995;31(1):9–13. [15] Carter J, Tomson C, Stevens P, Lamb E. Does urinary tract infection cause proteinuria or microalbuminuria?. A systematic review. Nephrol Dial Transplant 2006;21(11):3031–7. [16] Oke J, Stevens R, Gaitskell K, Farmer A. Establishing an evidence base for frequency of monitoring glycated haemoglobin levels in patients with type 2 diabetes:
314
[17]
[18]
[19]
[20]
diabetes research and clinical practice 99 (2013) 307–314
projections of effectiveness from a regression model. Diabetic Med 2012;29(2):266–71. Buclin T, Telenti A, Perera R, Csajka C, Furrer H, Aronson J, et al. Development and validation of decision rules to guide frequency of monitoring CD4 cell count in HIV-1 infection before starting antiretroviral therapy. PLoS One 2011;6(4):pe18578. Alleyn C, Volkening L, Wolfson J, Rodriguez-Ventura A, Wood J, Laffel L. Occurrence of microalbuminuria in young people with Type 1 diabetes: importance of age and diabetes duration. Diabetic Med 2010;27(5):532–7. The adolescent type 1 diabetes cardio-renal intervention trial research group. Adolescent type 1 Diabetes Cardiorenal Intervention Trial (AdDIT). BMC Pediatr 2009;9:p79. Hirst J, Taylor K, Stevens R, Blacklock C, Roberts N, Pugh C, et al. The impact of renin-angiotensin-aldosterone system inhibitors on Type 1 and Type 2 diabetic patients with and without early diabetic nephropathy. Kidney Int 2012;81(7):674–83.
[21] Dunger D, Schwarze C, Cooper J, Widmer B, Neil A, Shield J, et al. Can we identify adolescents at high risk for nephropathy before the development of microalbuminuria? Diabetic Med 2007;24(2):131–6. [22] Amin R, Widmer B, Prevost T, Schwarze P, Cooper J, Edge J, et al. Risk of microalbuminuria and progression to macroalbuminuria in a cohort with childhood onset type 1 diabetes: prospective observational study. BMJ (Clinical research ed) 2008;336(7646):697–701. [23] KDOQI. KDOQI Clinical practice guidelines and clinical practice recommendations for diabetes and chronic kidney disease. Am J Kidney Dis 2007; 49(Suppl. 2): p. S12-154. [24] Marcovecchio M, Dalton R, Turner C, Prevost T, Widmer B, Amin R, et al. Symmetric dimethylarginine, an endogenous marker of glomerular filtration rate, and the risk for microalbuminuria in young people with type 1 diabetes. Arch Dis Child 2010;95(2): 119–24.