Utility of Casual Postprandial Glucose Levels in Type 2 Diabetes Management

This study was conducted at the Grady Diabetes Clinic in Atlanta, Georgia, an outpatient facility that provides comprehensive diabetes care to inner-city residents. Most patients are African Americans and have type 2 diabetes (classified on the basis of typical clinical characteristics) (13). Patient characteristics available in the Grady Diabetes Patient Tracking System (DPTS) include date of birth, sex, race, date of diagnosis of diabetes, and type of therapy at presentation and at each subsequent visit. Patients were selected if they had type 2 diabetes, had an initial visit to the clinic between 1991 and 1998, had been followed in the Diabetes Clinic for at least 6 months (to exclude patients with new-onset diabetes in whom recent improvements in glucose control may not yet be reflected in their HbA_1c levels), and had a casual postprandial plasma glucose ([cPPG] 1–4 h postprandially; patients were asked when they last ate) and HbA_1c measured on the same day. For patients with multiple visits during the 1991–1998 period, only the last visit was selected. Based on the available data, we identified 1,827 unique patients having these characteristics.

The data were collected in part as baseline evaluation for the Improving Primary Care for African Americans with Diabetes (IPCAAD) study, a randomized, controlled trial to determine whether interventions aimed at provider behavior can improve diabetes control in a patient population at high risk for complications due to diabetes (14). The primary goal was to identify a cPPG cutoff that would identify close to 80% of patients with elevated HbA_1c (≥7.0% or >6.5%, indicating inadequate control by American Diabetes Association and American College of Endocrinology criteria, respectively).

Both plasma glucose and HbA_1c levels were measured at the Grady Memorial Hospital Laboratory. Plasma glucose levels were measured using a Hitachi 717 instrument (Roche, Indianapolis, IN). HbA_1c was determined with National Glycohemoglobin Standardization Program-certified instruments: a Diamat high-performance liquid chromatography system from Biorad (Hercules, CA) before 1997, and a Hitachi 717 immunoturbidimetric system from Roche (Indianapolis, IN) after 1997. Both assays had a normal range of 3.5–6.0% and were believed to provide comparable results (r = 0.98). Although a small number of patients in the present study had subnormal HbA_1c levels, such values usually cannot be explained on the basis of hemoglobinopathies and may reflect shortened survival of red cells. We used linear regression analysis to study the relationship between cPPG and HbA_1c. Receiver operating characteristic (ROC) analysis was used to study the sensitivity, specificity, and positive predictive value of specific cPPG cutoffs in predicting HbA_1c. ROC curves were generated using True Epistat statistical software (Richardson, TX), which also provides a calculated area under the ROC curve. The larger the area under the curve, the more accurate the test; an associated P value <0.05 was considered statistically significant.

The mean age (±SD) of the 1,827 patients was 57 ± 13 years, and mean duration of diabetes was 8.1 ± 8.0 years; 91% were African American, and 66% were women. The average HbA_1c was 8.4 ± 2.6%, and mean cPPG was 201 ± 94 mg/dl. Treatment consisted of diet alone (average HbA_1c 7.2%, n = 348), oral agents (HbA_1c 8.3%, n = 610), and insulin alone or in combination with oral agents (HbA_1c 9.1%, n = 869). The HbA_1c level was ≥7.0% in 67% of patient visits and >6.5% in 77% of visits.

Figure 1 shows that cPPG was significantly correlated with HbA_1c (r = 0.63, P < 0.001). Because patients treated with insulin were expected to have higher variability in their plasma glucose, we also studied the relationship between cPPG and HbA_1c according to type of therapy. The correlation between cPPG and HbA_1c was strongest in patients treated with diet alone (r = 0.75, P < 0.001). Even though the correlation between cPPG and HbA_1c was weaker in patients treated with oral agents (r = 0.64, P < 0.001) or insulin (r = 0.56, P < 0.001), it remained highly significant. Figure 2 shows the average (SE) HbA_1c for patients with cPPG in the ranges 76–125, 126–175, 176–225, 226–275, 276–325, and 326–375 mg/dl; an average cPPG of 105 mg/dl corresponded to HbA_1c of 7.6%, cPPG of 153 mg/dl corresponded to HbA_1c of 8.0%, and cPPG of 201 mg/dl corresponded to HbA_1c of 8.8%.

To be useful for health care providers in decision-making for intensification of therapy, a test should be both sensitive, to detect most patients with poor glycemic control, and specific, to exclude patients with good glycemic control. Our primary goal was to identify a cPPG cutoff that would have a sensitivity and a positive predictive value of ∼80%. Using ROC analysis, we found that cPPG was a significant predictor of an HbA_1c ≥7.0% and >6.5%, with an area under the ROC curve of 0.78 for both (P < 0.001 for both). A cPPG cutoff value of 150 mg/dl constituted a convenient indicator that predicted an HbA_1c level ≥7.0% in the whole group with a sensitivity of 78% and a specificity of 62%. The same cutoff predicted an HbA_1c level >6.5% with a sensitivity of 74% and a specificity of 66%.

Although the sensitivity and specificity of a cutoff provide useful information to identify good or poor glycemic control in a population of patients, positive predictive value is the main determinant of the utility of a cutoff with respect to individuals within the population. In this study population, in which the prevalence of an HbA_1c level ≥7.0% was 67% and the prevalence of HbA_1c >6.5% was 77%, a cPPG cutoff of 150 mg/dl had a positive predictive value of 80% in identifying an HbA_1c level ≥7.0% and a positive predictive value of 88% in identifying an HbA_1c level >6.5%. For each HbA_1c target, the positive predictive value was highest in insulin-treated patients and lowest in patients managed with diet alone (Table 1).

Our study shows that a cPPG level can be used to guide intensification of diabetes therapy in the absence of home glucose monitoring records or a current HbA_1c level. In our clinic, cPPG of 150 mg/dl identified 78% of those with HbA_1c ≥7.0 and 74% of patients with HbA_1c >6.5%. The selection of the cutoff of 150 mg/dl was based largely on the associated positive predictive values; the overlapping of the points in Fig. 1 obscures the fact that, in a population of patients with diabetes, a cPPG of 150 mg/dl obtained during a typical clinic visit often reflects a surprisingly high HbA_1c level (Fig. 2). The specificity of this cutoff was 62–66%, i.e., one-third of patients with adequate HbA_1c levels would be falsely identified as inadequately controlled, but the predictive value of the 150 mg/dl cutoff was 80–88%, i.e., 80–90% of patients with plasma glucose level >150 mg/dl also had an elevated HbA_1c. The cPPG was better correlated with HbA_1c in patients treated with diet alone or with oral agents as compared with patients treated with insulin (alone or in combination with oral agents). Although the present study was aimed at identifying a cutoff that could be used to guide practical patient management to achieve glycemic control, the findings are consistent with our previous examination of the utility of fasting and random plasma glucose levels in predicting poor glycemic control (15) as well as with previous observations of greater variability in plasma glucose values in insulin-treated patients (16).

Even though fasting plasma glucose has been used along with HbA_1c as a marker of glycemic control (17,18), many studies suggest that its utility is limited or that it is inferior to a nonfasting glucose. Avignon et al. (19) showed that, in patients with type 2 diabetes treated with diet alone or with oral agents, nonfasting plasma glucose was better than fasting plasma glucose in predicting glycemic control as reflected by HbA_1c. Even in insulin-treated patients with type 2 diabetes, the combination of prelunch and predinner blood glucose measurement seemed to reflect overall glycemic control better than the fasting glucose, as long as patients were on a stable insulin regimen (20). Information from the Diabetes Control and Complications Trial also showed that, in patients with type 1 diabetes, fasting plasma glucose tended to underestimate the HbA_1c level, whereas postmeal plasma glucose was a better indicator of HbA_1c and glycemic control, especially when measured after lunch (21).

The availability during clinic visits of a glycemic indicator that reflects HbA_1c would be a useful tool to guide intensification of diabetes therapy in inadequately controlled patients. Providers frequently believe they cannot evaluate a patient’s glycemic status unless they have a current HbA_1c level or records of the patient’s home blood glucose monitoring. A previous HbA_1c level would probably be of little help, because we have shown previously that when both a current plasma glucose and a 2-month-old HbA_1c measurement are available, health care providers tend to put more emphasis on the plasma glucose level in their clinical decision-making (22). Therefore, the use of a specific cutoff for a casual plasma glucose that identifies most inadequately controlled patients could prompt providers to advance pharmacologic therapy when needed and help minimize the problem of “clinical inertia”—failure to intensify therapy when indicated (23). Consistent with this hypothesis, we have found a higher likelihood of action by physicians (less clinical inertia), an increase in intensification of therapy, and corresponding improvement in HbA_1c levels when a cPPG of 180 mg/dl was used to prompt intensification of therapy (24).

Our study has some limitations. Because the specificity of a 150-mg/dl glucose cutoff is 62–66%, approximately one-third of patients with adequate HbA_1c levels would be falsely identified as poorly controlled and might have their therapy advanced, therefore putting them at risk of hypoglycemia. However, we have previously shown that hypoglycemia tends to be uncommon and mild in patients with type 2 diabetes, even in a setting in which intensive diabetes therapy is guided by treatment algorithms based on glucose levels obtained during office visits (25). Use of the 150-mg/dl guideline in the Grady Medical Clinic has not been believed to result in major problems with hypoglycemia, and we have found, in a retrospective analysis, that among patients who had an HbA_1c <6.5% and who had their pharmacologic therapy advanced, only 12% reported hypoglycemia on follow-up 3 months later (unpublished data). Nonetheless, clinical judgement should be exercised in decision-making. Therefore, physicians often see patients who present with high glucose levels but have had no recent HbA_1c levels and must choose between immediate action on the basis of guidelines that are useful but imperfect or delayed action via ordering repeat HbA_1c tests and follow-up appointments, which may or may not be kept. Because HbA_1c levels in Americans with diabetes have risen from an average of 7.8% in 1988–1994 to 8.1% in 1999–2000 (26), we favor ordering the test but also intensifying therapy if glucose levels are >150 mg/dl. Although it seems likely that capillary glucose (faster and less expensive) could be used instead of plasma glucose based on comparability in other settings (27,28), we did not test this hypothesis directly, and it is possible that cutoffs would need to be standardized to individual test systems (27,29). It is also possible that the relationship between cPPG and HbA_1c would be tighter if adjusted for time of day and hours since the last meal (30), along with differences in age, BMI, etc., but the resulting complexity (a nomogram of cutoffs) would likely make the guideline too cumbersome to use in practice. A similar “keep it simple” approach is generally preferred for screening for gestational diabetes (31,32).

Finally, it should be emphasized that the predictive values will be influenced by the prevalence of inadequate glycemic control in the patient population. As shown in our results, a better predictive value for identification of suboptimal glycemic control is obtained in subgroups in which poor glycemic control is more prevalent. Therefore, the cutoff plasma glucose of 150 mg/dl had better predictive value in patients treated with insulin or oral agents compared with patients managed with diet alone, because the patients treated with pharmacologic agents tended to have higher HbA_1c levels. Such observations suggest that the 150-mg/dl cPPG cutoff may be particularly useful to guide management in settings in which HbA_1c levels tend to be high, such as in some primary care sites (33–35).

In conclusion, we have identified a cutoff for casual plasma glucose that may be used as an indicator of glycemic control in patients with type 2 diabetes when home blood glucose monitoring records or current HbA_1c levels are not available. More frequent use of this marker to prompt intensification of therapy should help overcome “clinical inertia,” reduce HbA_1c levels, and improve diabetes outcomes. Although clinical judgment should still be exercised in the decision to advance therapy, expanded use of the 150-mg/dl glucose cutoff may be particularly important in the effort to improve diabetes management in the primary care setting.

This work was supported, in part, by research awards from AHRQ and NIDDK, #HS-07922 (L.S.P.).

This work was presented, in part, at the Annual Meeting of the American Diabetes Association, Philadelphia, 25 June 2001.