A Nonlinear Effect of Hyperglycemia and Current Cigarette Smoking Are Major Determinants of the Onset of Microalbuminuria in Type 1 Diabetes

Study protocols for baseline and follow-up examinations were approved by the Committee on Human Studies at the Joslin Diabetes Center.

Between 1 January 1991 and 31 March 1992, a 50% sample of patients aged 15–44 years attending the Internal Medicine or Pediatric Departments of the Joslin Clinic were screened for microalbuminuria in random urine specimens. Details of patient selection criteria and methods have been published previously (7,14,29), and an abbreviated description is included here. By 31 March 1992, 1,602 patients with type 1 diabetes had been screened at least once for microalbuminuria. During the next 2 years (the baseline interval), whenever they returned to the clinic for their routine care, their random urine specimens were examined for microalbuminuria. After 2 years had elapsed from the date of the first urine specimen, the nephropathy status of each patient at baseline was determined according to a consensus (defined below) of the available urine specimens. The 2-year baseline interval was selected because a majority of patients (61%) had at least three samples tested within that time. Of the 1,602 patients screened, 1,080 had normoalbuminuria. Of these, 943 patients provided follow-up urine specimens as described below. The remaining 137 (13%) provided no follow-up urines and were excluded from this analysis.

We used the ratio of urinary concentrations of albumin and creatinine (ACR; expressed in mg/g) for both the diagnosis of normoalbuminuria at baseline and the follow-up assessment of urinary albumin excretion. Urinary albumin concentration was measured by immunonephelometry (Behring, Somerville, NJ), and urinary creatinine concentration was measured by colorimetry (modified Jaffe reaction) on an Astra-7 automated system (Beckman Instruments, Brea, CA). Details of the laboratory methods have been published previously (7). Sex-specific criteria for normal urinary albumin excretion, as assessed by the ACR in a random urine, have been published previously: <17 mg/g for men and <25 mg/g for women (7). Higher values are considered to represent microalbuminuria (or proteinuria if ≥250 mg/g for men and ≥355 mg/g for women). During the initial phase of the study, urine samples with an albumin concentration <10 mg/l were deemed within the normal range and urinary creatinine was not determined (40% of the baseline samples). Subsequently, creatinine was determined in all urine samples regardless of the albumin concentration.

The diagnosis of normoalbuminuria during the 2-year baseline interval was based on a consensus of the available ACRs in that interval. For those with three or more measurements (61%), we required that >50% of the available specimens be in the normoalbuminuria range. For those with only two measurements (22%), we required that either both were in the normal range or, if one was abnormal, it was close enough to the normal range that the geometric mean of the two was within the normal range. Those with a single measurement (17%) were classified according to that single value.

For 4 years after the baseline period, we continued to measure the ACR in random urine samples collected from members of the cohort whenever they returned to the clinic. The 4-year period was divided for analytical purposes into 2-year intervals, designated as the first and second follow-up intervals. The distribution of patients according to the number of ACR measurements available during each follow-up interval remained fairly stable through both follow-up intervals (49 and 56% with three or more measurements, 25 and 21% with two, and 26 and 23% with one, for the first and second follow-up intervals, respectively). The same consensus criteria for determining a patient’s nephropathy status in the baseline interval were applied to the results in each follow-up interval. Persistent microalbuminuria was considered to have developed if the available ACR measurements in a follow-up interval failed to meet the criteria for normoalbuminuria. For patients with three or more measurements, this meant that half or more were abnormal; for those with two measurements, we required that both were abnormal or, if one was normal, it was close enough to abnormal that the geometric mean of the two was in the abnormal range, and for those with a single measurement during the interval, it was in the abnormal range.

Because we required that a diagnosis of microalbuminuria be confirmed in a second urine sample (hence the designation persistent microalbuminuria), we repeated the analyses in this report after excluding the minority of microalbuminuria onsets that were based on a single ACR. The results, which were not materially affected, are summarized in a footnote to Table 3.

Since 1988, GHb has been assayed by the Joslin Clinical Laboratory at the time of routine clinic visits. In January 1995, the assay method was changed from an electrophoretic method (Corning Medical and Scientific, Corning, NY) for measuring HbA₁ to an ion-exchange chromatographic method (Variant; Bio-Rad, Hercules, CA) for the subfraction HbA_1c. Both methods were run together for 4 months, and the correlation coefficient for duplicate determinations on the same specimens was 0.98. The HbA₁ values were converted to HbA_1c values by the following formula: HbA_1c = (HbA₁−0.14)/1.23 (14).

The temporal relation between exposure to a level of hyperglycemia and the onset of microalbuminuria is unknown. Therefore, to select the best measure of glycemic exposure, we examined the dosage-response relation between the onset of microalbuminuria in each follow-up interval and summary measures of HbA_1c during three specific time periods preceding that follow-up interval. For this purpose, HbA_1c values for each study patient were retrieved from the laboratory’s archived databases and grouped into the following time periods: 1) 3rd and 4th years preceding each follow-up interval; 2) 1st and 2nd years preceding each follow-up interval; and 3) a weighted average of the first two values representing the period 1st to the 4th years preceding each follow-up interval. The average number of measurements of HbA_1c within these time periods were 3.5, 4.3, and 7.8, respectively. For all individuals, the geometric mean of their HbA_1c measurement in each 2-year interval was considered an integrated measure of their exposure to hyperglycemia during that interval. The quartiles of the distributions of the geometric means of the HbA_1c for the 1st and 2nd and the 3rd and 4th years preceding the follow-up interval were almost identical, and the Spearman correlation was 0.69. As an integrated measure spanning the 4-year time period, the two measures were averaged to weight the contributions of each 2-year interval equally.

All of the available medical records at the Joslin Clinic for individuals included in the study were abstracted. Information abstracted separately for baseline and the first follow-up interval included the first three measurements of systolic and diastolic blood pressure and the first measurement of weight in the interval. The mean of the systolic and diastolic blood pressures from each interval were used in the analysis. Height was abstracted from medical record, and BMI was calculated as kg/m² for the baseline interval.

Information regarding smoking status during the baseline interval was gathered by questionnaire (supplemented by telephone interviews with nonrespondents). Participants were asked if they currently smoked or had smoked in the past and, if so, the ages at which they had started and stopped smoking. Current cigarette smoking was defined as smoking at any point in the baseline interval. Past smoking was defined as smoking at an earlier time but not smoking in the baseline interval. The remainder had never smoked. Smoking status was determined for 75% (n = 709) of the cohort. The 25% without information differed from those with smoking information in several ways. They were younger (26 ± 8 vs. 29 ± 8 years; P = 0.0001), had shorter duration of diabetes (11 ± 8 vs. 13 ± 8 years; P = 0.001), and had lower systolic blood pressure (116 ± 10 vs. 118 ± 12; P = 0.001).

Analyses of univariate and generalized linear models were performed in SAS (SAS 6.12 for Windows; SAS Institute, Cary, NC). Generalized additive models (GAMs) were fit using SPLUS (Version 4.5; StatSci, MathSoft, Seattle, WA). Comparisons of baseline values of continuous and categorical variables among patients with and without onset of microalbuminuria were carried out using t tests and χ² tests. Variations in the level of potential confounders by categorized levels of baseline HbA_1c were assessed using analysis of variance. To estimate the incidence rate of the onset of microalbuminuria in the entire cohort, a person was credited with 2 person-years of follow-up for each interval in which they remained normoalbuminuric, and 1 person-year in the interval in which they had the onset (29).

Univariate and multiple logistic regression analyses were performed using pooled logistic regression (30). Pooled logistic regression is essentially a discrete failure time approach and was suggested by the use of a consensus of multiple ACR measurements obtained over a 2-year interval to define onset of microalbuminuria, as opposed to a single measurement that would have a specific date. This means that individuals entered the logistic regression for follow-up one time from baseline to the first follow-up interval and a second time from the first to second follow-up intervals, provided they did not have an event during the first follow-up interval. One complication was the group of 76 individuals who had no ACR measurement in the first follow-up interval but did have measurements in the second follow-up interval. Random imputation was used to assign an interval to events that occurred in that group (31).

The model for multiple logistic regression analyses included as covariates HbA_1c, smoking status, age at diagnosis of diabetes, and duration of diabetes (32). As a consequence of the study design, all models also included an indicator variable for the follow-up interval of analysis (first or second). Baseline smoking status was used for both periods of analysis. Because of missing creatinine values for many baseline urine samples that had very low albumin levels (<10 mg/l; see above), models that included ACR as a covariate were run in two ways. The first method restricted the analysis to the second follow-up interval for which urine creatinine as well as albumin were measured for all urine samples. The second method replaced each missing baseline ACR value with an average ACR based on urine samples with the same albumin concentration that did have creatinine measured. The resulting odds ratios were similar. Only the later analysis with its much larger sample size is presented.

We used GAMs, within the discrete failure time approach, to assess the linearity (in the logarithm of odds scale) of the relationship between HbA_1c and onset of microalbuminuria. A GAM allows for relaxation of the assumption of linearity by using a flexible smoothing term—in this case the loess option—to fit a smooth curve to the ungrouped data (33). We examined a variety of spans and found that a span of 0.7 best represented the association, as judged by Akaike’s information criterion. The results are presented as predicted incidence rates, with the shaded region representing the pointwise ± 2 SE band. Interactions between HbA_1c and smoking status (current smokers versus past smokers/those who never smoked) were tested using linear and nonlinear terms for HbA_1c.

For calculation of adjusted incidence rates within various subgroups of patient characteristics, the confounding variables were set to their mean or observed proportions. The 2-year follow-up intervals were taken into account in estimating the risk of onset per year and the incidence rate per 100 person-years from the logistic and GAM models, respectively (32).

The cohort of eligible individuals for this study consisted of 1,080 Massachusetts residents ages 15–44 years with type 1 diabetes (94% of them Caucasian) who were classified as normoalbuminuric during the 2-year baseline interval. The average age at diagnosis of type 1 diabetes in the cohort was 15 ± 8 years, and the average duration of diabetes at baseline was 13 ± 8 years. Follow-up screening for microalbuminuria was obtained from 943 individuals (87%). Of those followed, 92% were screened in the first 2-year follow-up interval and 84% in the second 2-year follow-up interval, with a resulting 3,245 person-years of follow-up. Persistent microalbuminuria (hereafter referred to as microalbuminuria) developed in 109 individuals in either the first or second follow-up interval, giving an incidence rate of microalbuminuria of 3.3 per 100 person-years.

Baseline characteristics of the patients are summarized in Table 1 according to whether microalbuminuria developed during follow-up. Compared to those remaining normoalbuminuric, the patients with the onset of microalbuminuria had a longer duration of diabetes (2 years on average; P = 0.03) and a correspondingly younger age at diagnosis of diabetes (P = 0.02). The strongest associations were a higher HbA_1c in those with the onset of microalbuminuria (9.1 vs. 8.1%; P = 0.0001) and a higher prevalence of current cigarette smokers (48 vs. 21%; P = 0.001). Current age, blood pressure, BMI, and sex were similar in the two groups.

An important aim of this study was to explore the temporal relationship between exposure to a level of hyperglycemia, as reflected in HbA_1c levels, and the risk of onset of microalbuminuria. Toward that end, we examined three summary measures of HbA_1c that had different temporal relationships with each follow-up interval. The first measure was the geometric mean of HbA_1c values obtained during the 3rd and 4th year preceding each follow-up interval. The second measure was the geometric mean of HbA_1c values obtained during the 1st and 2nd year preceding each follow-up interval. The third measure was the average of the first two measures, and represents an integrated measure of the glycemic exposure during the 1st, 2nd, 3rd, and 4th years preceding each follow-up interval. Each measure of HbA_1c was divided approximately into quartiles using the ranges 4.3–7.2, 7.3–8.1, 8.2–9.0, and 9.1.–16.5%, which were based on the distribution of HbA_1c values during the initial baseline interval. Incidence rates of microalbuminuria were estimated for each quartile (Table 2).

The HbA_1c measure based on values obtained during the 3rd and 4th years preceding each follow-up interval did not separate the study group very well into a gradient of risk levels for the onset of microalbuminuria (Table 2). The incidence rates for the first three quartiles were identical, and only the highest quartile had an elevated incidence rate—about three times as high as the lower quartiles. The HbA_1c measure based on values obtained during the 1st and 2nd years preceding each follow-up interval gave a slightly higher incidence rate for the highest quartile (6.9 vs. 6.4 per 100 person-years) and a lower incidence rate for the lowest quartile (1.3 vs. 2.2 per 100 person-years) (Table 2). Therefore, HbA_1c values during the 1st and 2nd years appear to be better discriminators of risk than the HbA_1c measure during the 3rd and 4th years before follow-up.

When the two HbA_1c measures were combined to give a measure based on a larger number of HbA_1c determinations that spanned 4 instead of 2 years, the incidence rates were not better than those based on only the 1st and 2nd years before follow-up. Addition of the HbA_1c values from the earlier time period produced a slight dampening of the risk gradient rather than an accentuation of it, as would be expected if the effects of hyperglycemia were cumulative over the whole 4-year period. To compare the measurements directly, we restricted the analysis to people with data in both 2-year intervals. As judged by χ² values for the coefficients for the HbA_1c variables and Akaike’s information criterion, the model using just the most recent HbA_1c values was a better discriminator of the risk of microalbuminuria than the measure based on 4 years (see footnote to Table 2). Accordingly, in all subsequent analyses we used the geometric mean of the HbA_1c values during the 1st and 2nd years preceding the follow-up interval.

In preparation for evaluating the dosage-response relation between hyperglycemia and the risk of onset of microalbuminuria without the potentially confounding effects of other risk factors, we examined the association of the significant risk factors identified in Table 1 with the level of HbA_1c. Patients in the lowest quartile of baseline HbA_1c were, on average, 3 years older at the diagnosis of diabetes and had a duration of diabetes 2 years shorter than patients in the higher quartiles. Of greater importance was that the proportion currently smoking increased progressively from 14% in the 1st quartile to 41% in the 4th quartile of HbA_1c.

Current smoking, therefore, was associated with both the onset of microalbuminuria and the level of HbA_1c, so it is a confounder of univariate estimates of the effect of HbA_1c on the risk onset of microalbuminuria. To evaluate the independent effect of HbA_1c level, we used multiple logistic analysis of these variables in patients with information on both smoking and HbA_1c (Table 3). Adjustment for smoking, age at diagnosis of diabetes, and duration of diabetes reduced the odds ratio for the highest quartile of HbA_1c from 6.3 in the unadjusted analysis to 4.7 (a 25% reduction). Inclusion of smoking in the model was responsible for most of the decrease in the estimates of the effect of HbA_1c.

To examine the shape of the association between HbA_1c and onset of microalbuminuria in more detail, we used a GAM that allows relaxation of the assumption of linearity between HbA_1c exposure and the outcome (24). The model was adjusted for smoking, age at diagnosis of diabetes, and duration of diabetes. There was a significant nonlinear component to the fit (P = 0.04): the incidence rate of onset was relatively constant for HbA_1c values of ∼8% and then increased for HbA_1c values >8% (Fig. 1). The change in the slope between the lower and upper ranges of HbA_1c indicated that the effect of a proportionate difference in HbA_1c depended on whether the HbA_1c was above or below 8%. To illustrate this finding, we estimated incidence rates in two distinct parts of the curve: at HbA_1c values 7.2 and 8% in the lower range and 9 and 10% in the upper range. For each pair of HbA_1c values, the lower was 10% less than the higher. The incidence rates for HbA_1c values at 7.2 and 8% were 1.36 and 1.54 per 100 person-years, respectively. This gave a difference of 0.18 onsets per 100 person-years. In contrast, the incidence rates for HbA_1c 1 and 10% were 3.24 and 6.53 per 100 person-years, respectively. This gave a difference of 3.29 onsets per 100 person-years. These rate differences correspond to rate reductions of 12% (from 1.54 to 1.36) in the lower range and 50% (from 6.53 to 3.24) in the upper range.

The effect of smoking on risk of onset of microalbuminuria was evaluated in a similar way. In the univariate analysis shown in Table 3, current smokers had a substantially elevated relative odds in comparison with those who never smoked (3.3; 95% CI2.1–5.2), whereas past smokers did not (0.82; 95% CI 0.39–1.7). The incidence rate of onset of microalbuminuria (per 100 person-years) was 7.9 for current smokers, 1.8 for past smokers, and 2.2 for never smokers (P = 2.0 × 10⁻⁷). To evaluate the independent effect of smoking, we used the multiple logistic analysis summarized in Table 3. Adjustment for HbA_1c and other potential confounders reduced the odds ratio for current smoking from 3.8 in the univariate analysis to 3.1 (an 18% reduction). The decrease was mainly attributable to the inclusion of HbA_1c in the model. Interestingly, when we restricted our analysis to individuals with at least three ACR measurements (or two concordant measurements) in both the baseline and the follow-up interval, the adjusted odds ratios for the 2nd through 4th quartiles of HbA_1c relative to the 1st were unchanged from the results of the original analysis (see legend to Table 3). The same was true for the current smokers.

To examine the joint effects of cigarette smoking and HbA_1c on the risk of onset of microalbuminuria, we fitted a GAM to obtain predicted incidence rates stratified by smoking status (Fig. 2). The model included an interaction term between the effects of smoking and HbA_1c and, as before, was adjusted for age at diagnosis of diabetes and duration of diabetes. The estimated incidence rates for nonsmokers and smokers at an HbA_1c of 8% were 1.3 and 4.8 per 100 person-years, respectively, a difference of 3.5 onsets per 100 person-years for smoking. The estimated incidence rates at HbA_1c values 8 and 10% among nonsmokers were 1.3 and 3.4 per 100 person-years, respectively, a difference of 2.1 onsets per 100 person-years for the effect of the higher HbA_1c. Neither of the differences associated with a single exposure were very large, but the presence of both factors together was accompanied by a very high estimated incidence rate—15.0 onsets per 100 person-years. The risk associated with exposure to both factors was more than additive. When either was evaluated as a second exposure, it was associated with a much larger effect than it had as a single exposure. Among individuals with HbA_1c of 10%, the added risk for smokers was 11.6 onsets per 100 person-years. Alternatively, among smokers, the added risk for those with poor glycemic control (HbA_1c = 10%) was 10.2 onsets per 100 person-years. With due allowance for this being an observational study, these differences could be viewed as indications of the potential benefits of either smoking cessation or improved glycemic control among smokers with poor glycemic control (Fig. 2).

The baseline level of the ACR has often been included in models of the risk of microalbuminuria (8,12,15,18,28). Interpretation of the estimated effect of this variable is problematic for both biological and statistical reasons, as ACR levels are also the outcome of interest (34). Nonetheless, ACR is a potentially useful clinical predictor of microalbuminuria. To test our findings while simultaneously taking baseline ACR into consideration, we included it in the multiple logistic model that included HbA_1c, smoking, and other potential confounders. A doubling of baseline ACR increased the odds by a factor of 4.1 (95% CI 2.9–5.9). Estimates of the odds ratios for HbA_1c and current smoking in this model were reduced only slightly (15 and 5%, respectively), indicating that their effects were independent of baseline ACR.

We followed prospectively for 4 years a large cohort of patients with type 1 diabetes to elucidate the association of two risk factors, hyperglycemia and cigarette smoking, to the risk of onset of microalbuminuria. Particular attention was given to the relation between onset of microalbuminuria and HbA_1c levels at various times in the past.

We found that the risk of onset of persistent microalbuminuria in each 2-year follow-up interval was more closely related to the HbA_1c measurements during the preceding 2-year study interval than to the interval 3–4 years prior or the average over the whole 4-year interval. If the renal injury caused by hyperglycemia is retained over a long time, the accumulated injury would be more closely reflected in the HbA_1c values from the preceding 4 years than in the values from just the preceding 2 years. Therefore, our results suggest that, although exposure to significant hyperglycemia even for a period of 2 years increases the risk of microalbuminuria during the subsequent 2 years, this effect is subsequently attenuated. One possibility is that, in part, the renal injury induced by hyperglycemia is functional and not carried over from earlier exposures, although more long-term data are needed to evaluate the extent and speed with which the effect of a period of severe hyperglycemia is attenuated.

This finding is consistent with the results of the Diabetes Control and Complications Trial (DCCT) study (10). In the primary as well as in the secondary intervention cohorts, the effect of intensive diabetes treatment on reducing the risk of microalbuminuria was observed only after good glycemic control was maintained for 3 years. Curiously, the recently published results of the post-trial follow-up appear at first glance to be inconsistent with this observation, in that the difference in risk in the two treatment groups has persisted despite the convergence of their HbA_1c distributions (35). Microalbuminuria had developed in 11% of those formerly in the conventional treatment group (average HbA_1c 9.1%) and in 5% of those formerly in the intensive treatment group (average HbA_1c 7.1%) (35). It should be noted, however, that in this post-trial study, significant differences in HbA_1c levels between the two groups persisted during the first 3 years and became insignificant only during the 4th year. Furthermore, microalbuminuria was evaluated in a single specimen taken 3.5 years after the end of the trial, and it is unknown how soon after the trial the microalbuminuria developed or in what proportion of these cases the microalbuminuria was transient rather than sustained. Our results suggest that the higher risk of onset of microalbuminuria in the conventional treatment group would persist 2 years beyond the end of the trial and perhaps longer, taking into account the persisting differences in HbA_1c between the study groups. Only subsequently would the difference between the groups decrease in response to the convergence of their HbA_1c distributions during the first several years of post-trial follow-up.

The results of our study and the DCCT are evidence that the risk of onset of persistent microalbuminuria is substantially determined by the level of hyperglycemia during a 2- to 3-year interval (window of exposure) preceding the onset of microalbuminuria. This has implications for further studies as well as for patient care. The use of a baseline value (that becomes increasingly remote during follow-up) or a cumulative average of measurements (that includes remote values no longer relevant to the current risk) may not give an accurate representation of the dosage-response relationship. This conclusion, however, must be qualified, because no one has specifically examined the effect of more remote exposures.

The shape of the dosage-response relationship between levels of HbA_1c and the risk of onset of microalbuminuria in the current study is similar to that obtained in our previous cross-sectional study (14). The shift to a steeper slope in the upper ranges of HbA_1c indicates that the effect of a proportionate difference in HbA_1c depends on whether the HbA_1c is above or below 8%. In the illustration given in results, a 10% decrease in HbA_1c from 8 to 7.2% was associated with a 12% lower risk of onset, whereas a 10% decrease in HbA_1c from 10 to 9% was associated with a 50% lower risk of onset. From this, one may infer that the expected benefit of improved glycemic control will depend in a similar way on whether the starting point is above or below HbA_1c= 8%, as well as an individual’s constellation of other risk factors (such cigarette smoking, as suggested by the data in Fig. 2).

In contrast, according to the dosage-response relationship modeled by DCCT investigators for “sustained microalbuminuria,” the expected risk reduction from a 10% improvement in HbA_1c is virtually the same in the ranges below and above 8.0, 49, and 35% (19). Two possible reasons for the different results from the DCCT and our study should be considered. (It should be noted that our comparisons between the two studies are restricted to DCCT results for “sustained microalbuminuria” [based on two measurements] because it is closer to our end point [based on two or more measurements for 71% of cases] than their less reliable end point of “microalbuminuria” [based on one measurement] [19]).

First, the distribution of duration of diabetes was quite different in the two studies. As reported in the DCCT and by us, duration has an impact on the risk of microalbuminuria (7,10). It may also have an impact on the shape of the dosage-response relationship with HbA_1c. Although there is not enough power in our study to test differences in the shape according to duration, the magnitude of the change in slope at HbA_1c = 8% was greater for patients with diabetes duration ≥10 years than it was in those with duration <10 years (data not shown). Patients with ≥10 years duration constituted 57% of our study cohort, but only 19% of the DCCT cohort (28).

Second, the two studies used different measures of exposure to hyperglycemia. In our study, we evaluated several indexes of hyperglycemia defined by various exposure windows. We found the strongest association with the geometric mean HbA_1c during the exposure window 2 years preceding the follow-up interval in which the outcome was evaluated. In the DCCT, the onset of microalbuminuria was related to the mean of all quarterly measurements of HbA_1c, up to the annual examination at which the first abnormal test was obtained (10,19,28). Therefore, the exposure windows related to the annual examinations ranged from 1 to 9 years and averaged about 4 years. If one takes as valid our finding that the risk of microalbuminuria is more closely related to the most recent 2 years of exposure than to past exposures, then inclusion of the more remote HbA_1c values in the updated mean may have blurred the relationship between HbA_1c and risk in the DCCT study.

In addition to the HbA_1c level, current smoking was significantly associated with the risk of the onset of microalbuminuria in the current study. Although this finding is consistent with the results of several previous studies (12,26,27), we were able to obtain new detailed information regarding this association. Because smokers had poorer glycemic control than nonsmokers, we examined the effect of smoking after controlling for HbA_1c using logistic regression models and GAMs. In the former, the effect of smoking was reduced because of its association with poor glycemic control, but the odds ratio remained high and was highly significant statistically. In the GAMs, on the other hand, we demonstrated that the effect of smoking was enhanced in individuals with poor glycemic control. These results indicate that hyperglycemia (as reflected in a high HbA_1c value) and current smoking may impact the risk of microalbuminuria through different pathophysiological mechanisms.

Because this was an observational study, one must consider the possibility that a person’s psychological makeup could predispose her or him to an unhealthy lifestyle that introduces various risk factors for microalbuminuria besides cigarette smoking and poor glycemic control. Therefore, some of the effects attributed to these two factors in our analysis may have been attributable to unmeasured risk behaviors. However, a large effect attributable to unmeasured risk behaviors seems unlikely, as the risk for past smokers was similar to that for those who never smoked. Furthermore, this result may be interpreted as evidence that the smoking-related renal injury is functional rather than cumulative (36). If true, a risk reduction from smoking cessation would be immediate.

Based on our evidence that the effect of smoking on the risk of microalbuminuria is magnified in patients with poor glycemic control, it is not surprising that the DCCT investigators found almost identical risks for current smokers and nonsmokers in the intensive treatment group (2.7 and 2.9 onsets of microalbuminuria per 100 person-years, respectively) (28). In the conventional treatment group, which had an HbA_1c distribution similar to that of our cohort, the risk for smokers was slightly higher (but not significantly) than in nonsmokers (5.3 and 4.2 per 100 person-years, respectively) (28). This difference is small in comparison to that seen in our study. However, if many of the smokers in the conventional treatment group stopped smoking during the trial, their excess risk would have been diminished, as suggested by the results for past smokers in our study.

Baseline ACR is a strong predictor of the onset of microalbuminuria, as reported previously (8,12,15,18,28). Interestingly, estimates of the effects of current smoking and HbA_1c were unaffected by inclusion of baseline ACR in the model—an indication that their effects are independent of ACR’s association with the risk of onset. Thus, ACR values within the normal range may be related to genetic determinants of susceptibility to kidney disease rather than to a response to environmental exposures (37). We did not find any association between systolic or diastolic blood pressure and the onset of microalbuminuria. This negative finding must be tempered by our use of measurements made in the clinic, the inaccuracies of which may have diminished our power to detect association. Previous studies, however, have not found a consistent relationship between blood pressure and the risk of microalbuminuria (11,12,15,18,28).

In conclusion, we investigated the relationship of two exposures, cigarette smoking and hyperglycemia (HbA_1c), to the risk of onset of microalbuminuria. The incidence rate of microalbuminuria changes almost imperceptibly as HbA_1c increases up to 8% and then rises steeply as HbA_1c increases beyond 8%. The precise value where the slope changes has not been determined. Current (but not past) smoking increased the risk of microalbuminuria. The effect of the two exposures together is greater than the sum of their separate effects, suggesting some type of interaction. One implication from these results is that patients with type 1 diabetes who both smoke and have high HbA_1c levels could greatly benefit from eliminating at least one of the two risk factors.

This research was supported by National Institutes of Health (NIH) Grant RO1-DK41526 and the Joslin Diabetes Center. L.J.S. was supported by NIH training grant T32 DK07260-21/22.

We are indebted to the patients of the Joslin Clinic and the staff of the Joslin Diabetes Center, particularly the Reception Desk, Clinical Laboratory, Management Information Systems, and Medical Records Departments for their assistance and cooperation in conducting this study, and to the following members of the Section of Genetics and Epidemiology: F. Denry, M. Davidson, M. Pezzolesi, M. O’Keefe, K. Anderson, D. Butler, N. Castronuovo, M. Wantman, and E. Hart.