A Multinational Comparison of Complications Assessment in Type 1 Diabetes: The DiaMond Substudy of Complications (DiaComp) level 2

DiaComp is a substudy of the WHO DiaMond study of type 1 diabetes incidence, the methods of which have been described elsewhere (2). Briefly, the DiaMond study was a collaboration between the WHO, the University of Pittsburgh, and the National Public Health Institute, Helsinki, Finland, and included 100 centers in 50 countries. Participants had to have been diagnosed with type 1 diabetes at <15 years of age, on insulin therapy at the time of diagnosis, and a member of a defined community. Between 1990 and 1994, the investigators determined the incidence of type 1 diabetes in these centers, which ranged from 0.1/100,000 per year in China to 36.8/100,000 per year in Sardinia (2). The overall DiaComp study comprised 25 centers that expressed interest in the complications study from 18 countries/territories in Asia, Africa, Australia, Europe, and North and South America. Fourteen of these centers were population-based DiaMond centers, whereas 11 were longer-duration centers invited by the steering committee that were thought to have populations representative of the local childhood-onset type 1 diabetic population. Information on complications, health care practices, and other risk factors was collected at two levels, the first with survey-only data and the second with an expanded survey plus a physician’s examination.

This report (level 2) concerns an interviewer-administered survey of type 1 diabetic participants to ascertain physician-diagnosed complications, health care practices, and behavior, plus an examination to further ascertain complication status, as well as two major risk factors, glycemia and blood pressure control. Nine centers from the DiaMond study were included. In addition, a further five centers were specially recruited to provide long-duration experience and represented large major sources of care (Argentina and Brazil) or formal non-DiaMond registries (Israel, Sweden, and Pittsburgh, Pennsylvania). In addition to the DiaMond criterion of age at diagnosis <15 years, the inclusion criteria also consisted of a diabetes duration between 5.0 and 24.9 years. Fourteen centers in 12 countries reported 1,188 individuals to be eligible for interview and exam. Of these, 37 were later determined to be ineligible (1 died and 36 moved) leaving 1,151 eligible individuals, of whom 892 (77%) responded and are described in Table 1. Those who did not participate were similar in duration (9.3 vs. 9.0 years), age (19.3 vs. 18.0 years), and sex (47.2 vs. 53.4% female) to those who did.

The DiaComp survey questions included history of physician-diagnosed complications, health care behaviors (e.g., blood glucose testing, insulin administration, number of visits to physicians, etc.), smoking status and history, level of education completed, occupational history, height, weight, medication and insulin use, the Michigan Neuropathy Screening Instrument (MNSI) questionnaire (4), and the Rose Questionnaire for angina and claudication (5). For level 2, further information was collected on alcohol intake, family history of diabetes and cardiovascular disease, and reproductive history. Surveys were translated into the local language and then independently back translated to English to verify the translation. A manual of operations detailed how to administer the survey, and at least one investigator from each center had to be trained and certified in at least one of eight workshops.

Lifestyle and health care practices were measured by self-report, including history of smoking, disability (as perceived by the patient and attributed to diabetes), number of physician visits for diabetes management and eye care, self-monitoring of blood glucose (SMBG), and intensive insulin therapy (IIT). SMBG was defined as self-monitoring of blood glucose at least once per day. IIT was determined as greater than two insulin injections per day or on the insulin pump. Duration of diabetes was categorized into two groups: short duration described those with diabetes duration from 5 to 14 years and long duration described those with diabetes duration from 15 to 24 years.

The level 2 examination consisted of a number of procedures, including sitting blood pressure with a random zero sphygmomanometer using the Hypertension Detection and Follow-up Program (HDFP) protocol, having systolic/diastolic blood pressure ≥140/90 mmHg, or treatment with antihypertensive medication considered positive for hypertension. Height and weight were recorded in centimeters and kilograms, respectively. Detailed cardiovascular and neurologic histories were taken to supplement the self-reported data recorded from the survey. A cardiovascular exam was conducted, which recorded the presence of ankle edema, basal crepitations, and grading of the carotid, dorsalis pedis, and femoral pulses. In addition, distal symmetric polyneuropathy was assessed using the exam component of the MNSI, which included examination of the feet, ankle reflexes, and vibration perception at the great toe. Sensory testing using the 10-g monofilament was also applied to the dorsum of the great toe, while the patient was asked to respond yes or no to whether the filament was felt. Seven correct responses out of 10 applications was considered normal, one to seven correct responses indicated reduced sensation, and no correct responses indicated absent sensation. Visual acuity was measured by the Logmar eye chart, and the number of the letter size was recorded for each eye.

A nonfasting blood sample was taken by venipuncture or finger prick for measurement of HbA_1c using the DCA 2000 (Bayer, Tarrytown, NY). A nontimed urine sample was also taken at examination, and with which the level of urinary albumin was measured by dipstick (Mircal II) in milligrams per liter. Four levels were thus recorded: 1) negative, 2) 20 mg/l, 3) 50 mg/l, and 4) 100 mg/l. Microalbuminuria was classified as levels 3 or 4, i.e., 50 or 100 mg/l of albumin in the urine, respectively. Appropriate technology to determine an albumin-to-creatinine ratio in the field was not available at the start of data collection.

To maintain quality control for all examination procedures, a detailed manual of operations including instructions for each exam and laboratory procedure (e.g., recording blood pressure, performing DCA 2000 analysis for HbA_1c and urine dipstick analysis, recording height and weight, and conducting cardiovascular and neurologic exams) was distributed to all centers and presented (through training sessions) to leading investigators from each center at one (or more) of eight training workshops. To be eligible for participation, each center had to have at least one investigator from the center attend a training workshop. In addition, nine blinded, frozen control samples were sent to each center for both HbA_1c and Micral II dipstick testing. All centers produced test results within predetermined acceptable limits (seven or more samples [78%] being in target range) for both analytes, with five exceptions (two for HbA_1c and three for Micral II dipstick). Appropriate analyses were repeated without these specific centers’ data, and as similar results were found, the full dataset is reported.

Reported complications were defined as a self-report of physician diagnosis and included retinopathy and laser treatment, neuropathy, kidney problems related to diabetes and/or albumin in the urine (i.e., renal disease), high cholesterol and/or triglycerides, and macrovascular complications including myocardial infarction, stroke, peripheral vascular disease, amputation, and angina. The MNSI questionnaire was also used as a marker for neuropathy and consisted of 15 yes or no questions with seven or more answered positively indicating an increased probability of neuropathy (4). Examined complications were defined by the MNSI exam, with a score of two or more as positive for distal symmetric polyneuropathy and the measurement of ≥50 mg/l of albumin in the urine as positive for microalbuminuria. Hypertension was defined as systolic blood pressure (SBP) ≥140 mmHg or diastolic blood pressure (DBP) ≥90 mmHg or on antihypertensive medication, and angina was based on the examining physician’s diagnosis at the local center.

Spearman correlation coefficients were calculated between reported and measured complications overall and within each center. Mean SBP and HbA_1c differences were measured by complication status, and correlation coefficients were also calculated between center complication prevalence and the mean HbA_1c and SBP. Logistic regression was used for the individual data to model the effect of center on complication status. Each model contained indicator variables representing the centers (reference center was the center for which the number of participants most closely approximated the overall mean participants). For neuropathy, one center (Ancona) had a zero prevalence, so the analyses were repeated using exact logistic regression. The exact model did not diverge from the asymptotic (without Ancona), so the latter is reported for consistency with the other models. Subsequently seven models were then compared. The first model included only the center effects. The second included the center effects plus the demographic variables, duration and sex. The third included center effects, demographic variables, and HbA_1c. The fourth model included center effects, demographic variables, HbA_1c, and blood pressure (results were similar when hypertension was used). The fifth model included all of the above plus smoking status. The sixth model contained all previous variables plus microalbuminuria status. The last (full) model included all of the demographic and demonstrated risk factors listed above plus health care practice variables (IIT and SMBG).

The number of examined participants and their mean ages and durations are shown for each center and region in Table 1, along with the proportion who were female. There was no substantial geographic variation in mean age or mean duration. The distribution of sex was also fairly equal.

The distributions of smoking status, diabetes care practice variables (physician visits, daily blood glucose monitoring, and intensive insulin therapy), SBP, and HbA_1c are presented in Table 2. As a health-related behavior, smoking was generally consistently low in the short-duration group, with some exceptions (e.g., Romania and Argentina), and consistently higher in the long-duration group (Table 2). Physician visitation, as measured by either having seen a physician in the previous year or by the number of visits in the past year, showed generally good access to physicians for diabetes care, although Romania did stand out with markedly low visitation. Daily SMBG is important for proper glycemia management and showed a high degree of variation in both the long- and short-duration groups, with Romania, Puerto Rico, and Sweden particularly low. The two demonstrated risk factors, SBP and HbA_1c, were not widely varied in general, although HbA_1c did show some variation in the short-duration group, albeit moderately so.

The prevalence of complications also shows substantial variation for nearly all complications in both duration groups (Table 3). Reported renal disease, though generally low in the short-duration group, was high in three of the eastern European centers and in Puerto Rico, whereas reported prevalence in the long-duration group ranged from 21.4% (Pittsburgh, Pennsylvania) to 60.7% (Finland). Measured levels of microalbuminuria, however, showed no correlation with reported kidney disease or proteinuria (overall r = −0.05, P > 0.8) at the center level (but did correlate fairly well on an individual level, see below). As expected, there was substantially more measured microalbuminuria than reported for the short-duration group, but not the long-duration group. Microalbuminuria also showed considerable heterogeneity across center in both duration groups. On an individual basis, there were generally strong correlations between reported and measured rates both within center and overall. However, using microalbuminuria to validate reported renal disease, we found that, while each measure identified almost exactly the same number of positive cases in subjects (187 positive by self-report and 188 positive by Micral II dipstick), less than one-half (48%) of those who reported renal disease were also identified as being positive for microalbuminuria. However, it is also notable that 83% of those who reported no renal disease were indeed subsequently confirmed not to have microalbuminuria by Micral II testing.

Reported neuropathy again stood out in eastern Europe in the short-duration group, whereas in the long-duration group it was more consistent. The neuropathy exam correlated significantly with reported history at both the center (r = 0.5, P < 0.05) and individual (r = 0.28, P < 0.001) levels. The high self-reported results for eastern Europe (Lithuania and Romania) for the short-duration group were confirmed, with Puerto Rico also being identified as having a high (59%) prevalence. This high prevalence in Puerto Rico was to some degree linked to linea pedis, which is scored as an abnormality in the MNSI. For the long-duration group, the results of the exam showed similar patterns of variation to those of reported neuropathy except that, as expected, the exam identified higher prevalences for all centers. Neither reported nor measured hypertension was markedly varied for either duration group, though Finland did show twice the reported prevalence of other long-duration centers. Measured hypertension showed a strong correlation to reported rates both by center (r = 0.61, P < 0.01) and individually (r = 0.49, P < 0.001). Finally, κ-statistics were calculated for renal disease proteinuria (0.30), neuropathy (0.30), and hypertension (0.50).

Summaries of the multiple logistic regression models are presented in Table 4, where only the centers with a significant effect are presented. For microalbuminuria, with the exception of Pittsburgh, the center effect was only slightly reduced after controlling for demographic variables. However, after controlling for HbA_1c, the effect of center was substantially reduced for Puerto Rico (odds ratio [OR] reduced from 2.3 to 1.8, 95% CI 0.71–4.2), Romania (from 3.6 to 2.9, 1.3–6.8), and Buenos Aires (from 3.9 to 3.2, 1.1–8.8). While SBP was a significant predictor of microalbuminuria prevalence, adjustment for it did not further reduce center effects, except only moderately for Buenos Aires (from 3.2 to 2.8, 1.0–8.1). Likewise, although smoking was independently associated with increased prevalence of microalbuminuria, its inclusion in the model did not substantially reduce center effects any further. Controlling for health care practice variables (SMBG and IIT) did, however, contribute to the additional reduction of center effects for Puerto Rico (from 1.8 to 1.2, 0.42–3.7) and Romania (from 3.0 to 2.1, 0.86–5.4). These effects probably reflect SMBG, which showed a strong protective effect against microalbuminuria prevalence (0.59, 0.40–0.90), rather than IIT, which showed no clear protective effect (0.92, 0.52–1.6). Thus some of the geographic variation in microalbuminuria is explained by HbA_1c and SMBG, and hypertension further contributes prevalence prediction.

The center effect on examined neuropathy status was similarly assessed and was substantially reduced for all centers, except Puerto Rico, after controlling for the demographic variables (Table 5), so that only for Puerto Rico do the CIs then exclude unity. Continued attenuation of the center effect in Puerto Rico (OR reduced from 44.0 to 28.3) occurred after controlling for HbA_1c. However, controlling for microalbuminuria, SBP, smoking, and health care variables did not further explain the high Puerto Rican effect, although SBP (OR = 1.04, 95% CI 1.02–1.06) and microalbuminuria (2.0, 1.2–3.4) were both significantly positively associated with neuropathy status. Notably, the strong relationship between blood pressure and neuropathy did not appear to be mediated by microalbuminuria, as SBP remained independently associated with neuropathy after controlling for microalbuminuria.

The measured complications in DiaComp showed marked geographic variation, e.g., microalbuminuria (10-fold), neuropathy by exam (60-fold), and measured hypertension (48-fold). In this report, we were also able to directly compare the patterns of complication rates identified by both measures of each and assess their relationship and level of agreement. Generally, these two measures correlated well, e.g., for neuropathy and hypertension, although this was not true for renal disease/proteinuria on an intercenter basis.

Although reported renal disease and microalbuminuria both showed substantial geographic variation and they both showed high rates in eastern Europe, the two measures were otherwise divergent, showing little overall agreement (Table 3). This low correlation is apparent without many instances of marked incongruency (i.e., one measure indicates a center is low relative to the others, whereas the second measure indicates that that same center is high). So it would seem that the lack of association between the two measures may be more the result of one measure (self-reported renal disease) not detecting much of the disease that is actually there and/or the reported measure reflecting that the disease is no longer present. This was further assessed by the use of microalbuminuria prevalence to validate reported rates. The lack of sensitivity of the reported renal disease is likely to reflect both generally inadequate screening applied to patients with type 1 diabetes and variation in revealing these screening test results to the patients. Another issue is whether our reported measure “kidney problems” and/or “protein in the urine” is a fair indicator, e.g., the former may reflect minor past infection. However, 83% of those with “kidney problems” also reported proteinuria. The results were essentially the same if those with kidney problems but no proteinuria were excluded. Moreover, for renal disease/proteinuria in particular and all complications in general, the decreased sensitivity of self-report compared with exam was clearly more pronounced for the short-duration group than for the long-duration group (Table 3), which may be a further result of screening being differentially applied to those of short duration, perhaps because of scarce resources that must be discriminately applied to sicker people (who would generally have a longer duration of diabetes).

Neuropathy was high in the eastern European centers (Lithuania, Romania, and Martin) for the short-duration group and exceptionally high in Puerto Rico (short duration) by exam. For those of long duration, neuropathy was also high by either measure in Israel and Finland (Table 3). While it is to be expected that the prevalence of neuropathy by exam will be higher than by self-report, it is important to note that the same geographic patterns of variation were apparent in either measure, and, moreover, these measures were generally in agreement, showing high positive correlation within centers and overall. Hypertension also was high in eastern Europe (Slovakia) by either self-report or exam (albeit higher for exam) for the short duration and high, again, in Finland for the long duration (Table 3). In addition, although both measures again show substantial geographic variation, they also show a high level of agreement within centers and overall. Finally, neuropathy and hypertension both show agreement between self-reported and directly measured rates; though based on the κ-statistics, it would appear that the agreement could only be called moderate for hypertension (7). On the other hand, self-reported and directly measured rates of renal disease/proteinuria show a distinct lack of agreement based on the low-correlation incongruent centers, but interestingly, a κ-statistic (0.3) similar to neuropathy.

Glycemia, surprisingly, showed only a modest association with complications prevalence. It should be noted, however, that controlling for HbA_1c did reduce the center effect for Puerto Rico, Romania, and Buenos Aires for microalbuminuria. The observed relationships with HbA_1c may be due to the fact that glycemia was generally high across many centers. Another possible explanation is that one HbA_1c measurement is unlikely to reflect glycemia control over the long term. However, on a relative scale, this observation should be seen in the context of the SBP observations, which were also based on one examination and showed a stronger association despite greater day-to-day variability. We must conclude, therefore, that factors beyond glycemia are contributing to variation in complication prevalence, though this does not diminish the established role of glycemia in the pathogenesis of microalbuminuria and neuropathy.

SBP was strongly and significantly positively associated with both examined neuropathy and microalbuminuria status, which is consistent with many other studies that show hypertension to be related to nephropathy (6) and neuropathy (8). The demonstrated association between blood pressure and neuropathy remained after controlling for microalbuminuria, a potential confounder of this relationship, even though microalbuminuria was itself strongly associated (OR 2.0) with neuropathy status. Although this was the case, SBP nevertheless failed to substantially contribute toward reducing the effect of center on complication status (Table 4). Health care practices and behaviors, including smoking, SMBG, and IIT, varied markedly across regions in both duration groups. Smoking was significantly associated with microalbuminuria, but not examined neuropathy, and contributed little toward reducing the effects of centers for either of these complications. Daily SMBG was strongly protective against microalbuminuria. In addition, IIT was generally protective against complication status, albeit more moderate and nonsignificant, but did not contribute toward reducing the center effects. Taken together, demonstrated risk factors and health care practices (SMBG) largely explained the variation observed for microalbuminuria status. However, while full explanation of the high rate of examined neuropathy seen in Puerto Rico remains elusive, the remaining high rates are explained by demographic variables. We conclude, therefore, that while SBP, glycemia, and health care practices could not account for all the geographic variation in the examined complications, they did explain much of the variation documented, especially for microalbuminuria.

The EURODIAB IDDM Complications Study, which was also a cross-sectional study of type 1 diabetes complications in 31 clinics in 16 European countries, showed more moderate geographic variation in microvascular complications than DiaComp. However, EURODIAB did show a substantial preponderance in eastern Europe (mean duration of 14.2 years) of neuropathy at 46.4%, compared with northwestern Europe (mean duration of 15.4 years) at 24.9% (9). On the other hand, EURODIAB did not demonstrate as great a degree of variation for microalbuminuria, as did DiaComp, but they did find that it was high (25%) overall for individuals of very short duration (≤5 years) (10), whereas DiaComp also showed microalbuminuria to be quite high (55%) in those of short duration (5–9 years) (Table 3). In addition, one center in EURODIAB also demonstrated blood pressure to be closely associated with microalbuminuria (as did DiaComp), but found that HbA_1c was not correlated with the degree of renal disease (excluding renal failure) (11), which, while generally contradicting the literature, supports the findings between HbA_1c and microalbuminuria in DiaComp.

For this (level 2) component of DiaComp, complication prevalence was assessed by both recall of physician diagnoses and by a physician’s examination. One aim of this report was to compare the two methodologies to see if similar patterns of rates could be identified by examination. In fact, we have shown that similar patterns were identified by both self-report and examination; however, certain potential biases do exist for each of these measures. The first measure requires five key steps. First, a physician must diagnose the condition. Second, the physician must inform the patient of the diagnosis. Third, the individual must understand and retain the diagnosis information. Fourth, the individual must be able to recall the diagnosis over time. Finally, the individual must be willing to communicate this diagnosis with others. Clearly self-reported diagnoses of complications are potentially limited by failure or bias in any of the above stages, which may vary across centers. For example, centers may vary in active screening for urinary albumin or in the threshold used to inform patients (e.g., microalbuminuria versus frank proteinuria). The second measure (i.e., examination) is certainly more complete and avoids the problems above. However, the potential for measurement error in exam and laboratory procedures in widely varying contexts across many different countries is also something to give due consideration. We have, however, applied strict quality control by training the investigators personally in the exam and laboratory protocol to be followed, as well as presenting each of them with a detailed manual of operations for each of the procedures and conducting analyses with control samples for HbA_1c and urine albumin measurement.

Although the DiaComp study has identified important patterns in type 1 diabetes complications and has explained some of the geographic variation present, the design of the study is limited by its cross-sectional nature, and therefore no causal inferences can be drawn. Clearly, the need for longitudinal data in geographically varied populations is now not only warranted, but critical, for a complete picture on the global occurrence of complications in type 1 diabetes.

Rosario, Argentina: A. Libman (PI). Buenos Aires, Argentina: O. Ramos (PI), M. Ferraro. Australia: D. McCarty (PI). Rio, Brazil: M. Gomes (PI). Sao Paulo, Brazil: L. Franco (PI), A. Ramos, C. Negrato, H. Almeida, M. Coral. Chile: E. Carrasco (PI). China: Yang-Ze (PI). Cuba: D. Diaz (PI). Israel: O. Kalter-Leibovici (PI), R. Berger, M. Karp, S. Shalitin. Ancona, Italy: V. Cherubini (PI), A. Iannilli, A. Gentili, A. Pinelli, S. Francolini, A.M. Paparusso. Chieti, Italy: F. Chiarelli (PI), S. Tumini, M. Cerruto. Japan (nationwide): N. Tajima (PI), Y. Uchigata (co-PI), K. Asao, R. Nishimura. Osaka, Japan: T. Kawamura (PI). Libya: Issam Hajjaji (PI). Lithuania: R. Zalinkevicius (PI). Romania: C. Ionescu-Tirgoviste (PI). Martin, Slovakia: J. Javorkova (PI), K. Javorka (PI). Sweden: G. Stenström (PI), B. Berger (PI). Oxford, U.K.: D. Dunger (PI), J. Edge, K. Ong, S. Strang. Chicago, IL: R. Lipton (PI). Pittsburgh, PA: T. Orchard (PI), J. Zgibor, M. Walsh. Puerto Rico: T. Frazer (PI), J. Vega, E. Barranco.

We thank Novo Nordisk for financial and administrative support.

This study was made possible by the volunteer efforts of the investigators listed in the appendix and their colleagues, diabetes nurses, and patients, for which we are deeply grateful. We also thank Drs. S. Kelsey and T. Songer for statistical advice and Drs. P. Bennett, L. Franco, and N. Tajima for serving on the steering committee.