A Prospective Population- Based Study of Microalbuminuria as a ...

O

R

I

G

I

N

A

L

A R T I C L E

C

A Prospective Population- Based Study of Microalbuminuria as a Predictor of Mortality in NIDDM ANDREW NEIL, MRCP MICHAEL HAWKINS, DPHIL MICHAEL POTOK, PHD

MARGARET THOROGOOD, PHD DAVID COHEN, MRCP JIM MANN, DM

OBJECTIVt — To assess prospectively the relationship between microalbuminuria and mortality in a geographically denned population of NIDDM patients and to determine the relative importance of microalbuminuria as a risk factor for mortality. RIS1ARCH DESIGN AND METHODS — A survey of known diabetes undertaken in 1982 identified a cohort of 249 NIDDM patients. Follow-up information was available for 246 patients who contributed 1498 person-yr exposure and were followed up for a mean period of 6.1 yr. The median age of the cohort at entry was 68 yr (range 28-89 yr), and the median duration of diabetes was 7 yr (range 1-41 yr). At baseline, a clinical examination was performed and a random daytime urine specimen was obtained for measurement of urinary albumin concentration. RESULTS — UAC results were available for 236 patients: 45 (19%)patients had a UAC > 15- 200 mg/L; and 145 (61%) had a normal UAC 7.8 mM. We identified 249 patients who had NIDDM, defined as diagnosed diabetes that either was treated without insulin, or if insulintreated, had been diagnosed after 30 yr of age in the absence of ketosis, and at least 1 mo had elapsed before the start of insulin treatment. Follow-up information was available for 246 of 249 NIDDM patients. Patients were interviewed and examined at baseline. Weight and height were measured in indoor clothing without shoes, and BMI (kg/m2) was calculated. Arterial BP (diastolic 5th phase) was measured with a standard clinical mercury sphygmomanometer on the right arm after a 10-min rest. Two BP measurements were made with a 1-min interval between, and the mean value was recorded. Hypertension was defined as untreated sBP >160 mmHg and dBP >95 mmHg, or as treatment of hypertension. Angina, possible myocardial infarction, and intermittent claudication were assessed with the WHO cardiovascular questionnaire (14), and CHD was defined by a positive response to the questionnaire for angina or possible myocardial infarction. The eyes were examined at entry to the study without knowledge of UAC. The presence of lens opacity in either eye was determined using direct ophthalmoscopy, after mydriasis, with the ophthalmoscope focused on the lens from — 150 mm. Aphakic subjects were recorded as having a lens opacity. Retinopathy was graded according to the Hammersmith system (15). Smokers were defined as patients currently smoking >1 cigarette/day. A random daytime urine specimen was obtained and the sample was excluded if the bacterial count exceeded 105 bacteria/ml (8 patients with bacteriuria and 2 with urinary catheters were excluded). Samples were stored at —20°C in polystyrene tubes, and the albumin concentration was determined

DIABETES CARE, VOLUME 16,

NUMBER 7, JULY

1993

within 2 mo with a double antibody R1A (inter batch CV 5.5% at 23 mg/L) (16). Serum total cholesterol was measured with an enzymatic method (kit no. 236691, Boehringer Mannheim, Lewis, UK,) using a Technicon A All continuous flow analyzer (Technicon Instruments, Basingstoke, Hants, UK). HbAlcwas measured with electrophoresis (Corning Medical, Halstead, Essex, UK, normal range 5.0-7.5%). The records of patients were flagged at baseline by the NHSCR so that we were automatically informed of deaths. The underlying cause of death was coded to the ICD (9 th revision) from copies of death certificates provided by NHSCR. Use of this national notification system enabled us to assume notification of all deaths from baseline until 30 June 1990 (17). By then the cohort had been followed up for 1498 person-yr, and the vital status (and cause of death where applicable) of 246 patients was known. Statistical analysis Results for univariate analyses are means ± SD for normally distributed continuous variables. UAC was log10 transformed because of a positively skewed distribution, and the results are stated as the geometric mean and 95% CI. We used one-way ANOVA to examine associations between categories of UAC and continuously distributed variables. Pearson's x 2 was used for dichotomous variables. All tests of statistical significance were two-tailed. Person-year cohort analysis was used to calculate the ratio of the number of deaths observed to the number expected using a computer program (18) that applies standard methods (19). Expected numbers of deaths were calculated by applying 5-yr age, sex, and single calendar year-specific death rates for the reference population resident within the boundaries of the Oxford Regional Health Authority. The ratio of the number of deaths observed to the number expected was expressed as a percentage (SMR). Exact 95% CI for the SMR was

calculated assuming a Poisson distribution for the observed frequency in the numerator. The SMR was calculated for all-causes mortality with four levels of UAC (see below). The relative importance of each variable in determining the time to death was assessed initially with the log-rank test of heterogeneity, stratified by age at entry, to compare survival estimates calculated from the life-table analysis (19). The baseline variables examined as possible confounders of any relationship between albumin concentration and time to death were the age at the time of examination, sex, known duration of diabetes, treatment (categorized into three levels: no treatment or diet; sulfonylurea and/or metformin; and insulin ), sBP, dBP, hypertension status, BMI, HbAlc, total cholesterol concentration, retinopathy, lens opacity, intermittent claudication, CHD, stroke, and current cigarette smoking. Continuously distributed variables, excluding UAC, were categorized into four groups by quartile value. Retinopathy was categorized as absent, < grade B, or ^ grade B, according to the Hammersmith grading system (15). UAC was classified into four groups: 15-200. A value of 15 mg/L discriminated between normal and elevated or borderline elevated UAC, and 200 mg/L was the upper limit of microalbuminuria (20). The use of these categories also enabled comparison with other published survival data (10). Statistical modeling with the Cox proportional hazard model was done with EGRET statistical software (21). This allowed us to assess the independent effect of UAC on mortality after adjusting for potentially confounding variables, identified by the stratified log-rank analysis. We examined potential interactions between UAC and confounding variables that might affect the time to death. UAC was entered into the model after classification into four groups as above. We entered all variables signifi-

997

Microalbuminuria

and mortality in NIDDM

Table 1—Clinical characteristics of patients (whose vital status was known), including those stratified by VAC UAC (MG/L)

n Adv. (YR) A G I - AT DIAGNOSIS ( Y R ) KNOWN DIABETES DURATION (YR) sBP

(MMHG)

dBP

(MMHG)

BM1

(KG/M 2 )

HbAu (%) TOTAL CHOLESTEROL (MM) HYPERTENSION (NO/YES)

RETINOPATHY ( 0 / < B / > B ) * LENS OPACITY (NO/YES) INTERMITTENT CLAUDICATION (NO/YES) CHD

(NO/YES)

STROKE (NO/YES) CIGARETTE SMOKING (NO/YES) SEX

(M/F)

N O TREATMENT OR DIET/SULFONYLUREA AND OR

15-200

P VALUE

145 65.6 ± 10.7 56.8 ± 11.1 8.3 ± 6.1 155 ± 2 1 87 ± 12 25.8 ± 3.9 9.8 ± 2.3 5.3 ± 1.1 84/61 114/18/13 94/51 135/10 124/21 140/5 122/23 77/68 40/75/30

45 64.6 ± 12.7 54.3 ± 13.6 9.9 ± 8.0 164 ± 28 91 ± 14 26.2 ± 9.4 10.1 ± 3 . 0 5.4 ± 1.1 19/26 33/8/4 23/22 41/4 36/9 43/2 36/9 20/25 11/21/13

36 68.7 ± 10.9 58.2 ± 13.0 0.2 ± 9.5 169 ± 22 90 ± 13 27.1 ± 11.2 10.4 ± 2 . 2 5.5 ± 0.9 16/20 26/5/5 20/16 29/7 32/4 35/1 24/11 22/14 6/26/4

10 69.8 ± 9.9 59.2 ± 7.6 8.2 ± 3 . 7 177 ± 16 95 ± 13 28.7 ± 4.2 9.7 ± 2 . 7 5.6 ± 0.9 4/6 4/4/2 6/4 7/3 9/1 10/0 8/2 3/7 1/7/2

NS NS NS 0.0002 NS NS NS NS NS NS NS NS NS NS NS NS NS

246 66.2 ± 11.3 56.6 ± 11.9 9.1 ± 7.3 159 ± 23 89 ± 12 26.2 ± 4.2 10.0 ± 2.5 5.4 ± 1.1 127/119 183/37/26 144/102 221/25 201/35 236/10 200/46 125/121 58/134/54

METFORMIN/INSULIN

Data are means ± SD. * Graded according to Hammersmith grading system (15).

cant in the age-stratified log-rank test into a stepwise forward selection procedure that included covariates that reached the highest level of significance at each stage using the score test until no further covariates were significant at the 5% level. After identifying the final model by this procedure, all variables that were not significant in the log-rank test were used separately to extend the model. We conducted a score test to determine whether significant additional variation had been explained. We checked the assumption of proportional hazards in two ways: by plotting the values of the logarithm of the cumulative hazard function for different values of the covariate to confirm (or refute) that these were reasonably constantly separated across time; also by fitting various interaction terms between the covariates and time (and its logarithm) to identify obvious departures from proportional hazards. We computed survival curves for each prognostic

998

variable after adjustment to the median value of the other variables in the final Cox regression model. RESULTS — Table 1 shows the clinical and biochemical features at entry to the study of the 246 NIDDM patients (121 men, 125 women) for whom follow-up information on vital status was available. The median age was 68 yr (range 28-89 yr) and the median duration of diabetes was 7 yr (range 1-41 yr). All except 6 patients were of European origin, and no significant differences in clinical features was evident between men and women. The results of a stratified univariate analysis of the relation between UAC (categorized in four groups) and other variables also are shown in Table 1 for 236 patients, after excluding those with urinary tract infections or catheters. The only significant association with albumin concentration was sBP (P = 0.0002). The geometric mean UAC in men was 12.8 mg/L (95% CI 10.3-15.9) and in

women 13.2 mg/L (95% CI 10.2-17.1). A UAC of > 15 - 200 mg/L was present in 81 patients (34%, 42 men and 39 women) and a UAC of 40-200 mg/L in 36 patients (15%, 22 men and 14 women). The 246 patients included in a person-year cohort analysis contributed 1498 person-yr at risk, and the mean follow-up period was 6.1 yr. Of the 93 deaths observed, 28 were attributable to CHD (ICD codes 410-414), 10 to cerebrovascular disease (ICD codes 430438), 18 to neoplasms (ICD codes 140239), 10 to respiratory diseases (ICD codes 460-519), and 27 deaths to other causes, including two from renal failure. Table 2 presents the SMRs for all-causes mortality and ischemic heart disease for both men and women, and for both sexes combined. The all-causes and CHD SMRs were significantly increased in women but not in men. No significant increase in mortality from cerebrovascular disease (ICD codes 4 3 0 - 4 3 8 ,


NUMBER 7, JULY

1993

Neil and Associates

Table 2—Person-years mortality analysis: all-causes and CHD PERSON-YR OF

OBSERVED

EXPECTED

OBSERVATION

DEATHS

DEATHS

SMR

95% Cl

P VALUE

782 716 1498

43 50 93

36.5 25.7 62.3

118 194 149

85-159 144-256 121-184

0.32 200 tig/ml 1000

1500 Time (dayB)

2000

2500

1500 iw (days)

3000

Figure 2—Survival curves for each of the four levels of UAC after adjustment to the median value of other variables in the model.

Figure 3—Survival curves for the presence or absence of lens opacity after adjustment to the median value of other variables in the model.

with a UAC of 16-29 mg/L also was reported in that study (8); we found a nonsignificant^ increased SMR of 156 in patients with a UAC >15- GRADE B

20 • 58

yre

53 • 67 ( r s 63 • 74

yrs

Z 75 jffs 1

2000 Tims (days)

Figure 1—Survival curves for each age-group after adjustments to the median values of other variables in the model.

1000

LENS OPACITY UAC

(MG/L)

>15-200

* Graded according to the Hammersmith grading system (15). Variables entered into the model: age, known duration of diabetes, retinopathy, lens opacity, intermittent claudication, and UAC.


NUMBER 7, JULY

1993

Neil and Associates

followed up for 3.5 yr (11). The estimate of risk, however, is necessarily imprecise because only 14 deaths occurred (95% CI 2.5-25.3), and the RR was reduced to i I 2.95 (95% CI 1.3-6.7) with continued follow-up at 6 yr (M.B. Mattock, unpubI lished observations). Comparison of our findings with those from clinic-based studies (4,10,11) that used statistical modeling to determine the impact of microalbuminuria on mortality is complicated by differences in Figure 4—Survival curves for retinopaihy the variables and how they were represtatus after adjustment to the median value of sented in different models. Although other variables in the model. studies have shown microalbuminuria consistently predicts increased mortality, the relative importance of most other collections have been shown to have a prognostic factors has varied. Age was a significant predictor in two studies CV as high as 38% (28). Our findings suggest a UAC of (4,10) and, when entered as a continu40-200 mg/L predicts a similarly in- ous variable in our analysis, we found a creased mortality in NIDDM, regardless similar RR of 1.1 for a 1-yr increase in of the untimed collection procedure age, compared with 1.07 in the largest used. In an earlier study of 492 NIDDM cohort (10). Duration of diabetes was patients followed for 10 yr (10), a similar also a significant predictor in that cohort analysis using a Cox's proportional haz- (10), but not in our study, nor in others ard regression model showed that a UAC (4,11). of 40-200 mg/L measured on an early Hypercholesterolemia was signifmorning specimen was associated with icantly associated with all-causes mortalan increased RR of mortality of 2.28, and ity in one study (11), but cigarette smoka UAC >15-