Higher Medical Care Costs Accompany Impaired Fasting Glucose

Study subjects were members of Kaiser Permanente Northwest (KPNW), a not-for-profit, group-model HMO. KPNW recommends lipid screening for men aged ≥35 and women aged ≥45. FPG tests are routinely ordered with lipid panels. Between 1 January 1994 and 31 December 2003, the KPNW Regional Laboratory analyzed 603,486 FPG tests for 231,093 unique individuals. Of the 113,687 patients who had at least two tests, we identified 28,335 patients with two or more FPG test results of at least 100 mg/dl and no evidence of diabetes (chart diagnosis of 250.xx, FPG >125 mg/dl, or use of an antihyperglycemic drug). We categorized these patients into two abnormal glucose levels: stage 1 (100–109 mg/dl) and stage 2 (110–125 mg/dl). The date of the earliest test in the stage was defined as the index date for that stage. Patients were followed until an FPG greater than the maximum of their current stage was recorded, an antihyperglycemic drug was dispensed, health plan membership terminated, or 31 December 2003. Most patients remained in a single stage throughout the study, but 3,281 subjects progressed from stage 1 to stage 2.

We matched each of the unique 28,335 subjects to a KPNW member who had a normal FPG test (<100 mg/dl) based on age, sex, and year of index FPG test. The resulting control subjects were followed until FPG ≥100 mg/dl was recorded, an antihyperglycemic drug was dispensed, health plan membership was terminated, or 31 December 2003. We further divided the 26,309 control subjects who had not died by the end of 2003 into those who had no FPG test ≥100 mg/dl (n = 23,621), had progressed to stage 1 FPG (n = 1,741), had progressed to stage 2 FPG (n = 462), or had progressed to diabetes (n = 485).

KPNW maintains electronic databases containing information on all inpatient admissions, pharmacy dispenses, outpatient visits, and laboratory tests. These databases are linked through the unique health record number that each member receives when they first enroll in the health plan. We recently detailed our determination of unit costs based on these databases (9). Briefly, cost coefficients were applied uniformly in all years, inflated to 2003 dollars. To minimize the effects of censoring, we annualized costs by dividing by months of observation and then multiplying by 12. We adjusted costs for age, sex, and comorbidities (history of myocardial infarction or stroke, depression, other CVD, or congestive heart failure) and weighted them by months of observation (to further account for censoring) using SAS Proc GLM version 6.12 (SAS Institute, Cary, NC).

The electronic medical record contains up to 20 physician-recorded ICD-9-CM diagnoses at each contact. Using these diagnoses, we identified comorbidities present at the time of the index FPG test. Smoking history, height, weight, and blood pressure were obtained from the electronic medical record. Lipid values were extracted from the laboratory database; for this study, we used the values nearest but before the index FPG test. We also counted the number of Adult Treatment Panel III criteria for the metabolic syndrome (10); however, because waist circumference was not available and FPG was the primary classification variable, the maximum number of criteria possible was three (triglyceride level ≥150 mg/dl, HDL level <40 mg/dl in men or <50 mg/dl in women, systolic blood pressure ≥130 mmHg, or diastolic blood pressure ≥85 mmHg).

To ascertain the independent contribution of FPG to annual total costs, we estimated a series of three regression models. The first included only FPG as the explanatory variable, the second added risk factors (age, sex, smoking history, BMI, systolic blood pressure, HDL cholesterol, triglycerides, and LDL cholesterol), and the third added baseline diagnoses (history of myocardial infarction or stroke, depression, other CVD, or congestive heart failure).

Subjects averaged 58.6 years old, and 54.0% were women (Table 1). Those with stage 2 FPG were ∼1 year older on average than stage 1 subjects (59.5 vs. 58.3, P < 0.001). CVD comorbidities of all types were much more common in the elevated FPG groups and were slightly higher in subjects with stage 2 FPG compared with subjects with stage 1 FPG. Use of antihypertensive, lipid-lowering, and other CVD drugs was also higher in subjects with elevated FPG stages and was greatest in stage 2 subjects.

We observed a graded difference in systolic blood pressure, HDL cholesterol level, triglyceride level, and BMI across the three groups, with more favorable mean values for subjects with normal FPG, and the poorest mean values for subjects with stage 2 FPG (P < 0.001 for all comparisons). Of the three possible criteria for metabolic syndrome (blood pressure ≥130/85 mmHg, HDL level <40 for men or <50 for women, and triglyceride level ≥150 mg/dl), 23.7% of subjects with normal FPG had none of them, compared with 12.8% of subjects with stage 1 FPG, and 8.1% of subjects with stage 2 FPG (P < 0.001). Conversely, 62.3% of subjects with stage 2 FPG had two or three criteria compared with 37.2% of subjects with normal FPG.

Age- and sex-adjusted annualized total costs (Fig. 1A) were lowest among those with normal FPG ($4,357), ∼$260 higher in subjects with stage 1 IFG ($4,617), and >$600 higher in subjects with stage 2 IFG ($4,966, P < 0.001 for all comparisons). Differences between the inpatient, outpatient, and pharmacy components of total costs, displayed in the stacks of the bars, rose similarly with FPG stage and were statistically significantly different in all between-group comparisons. After further adjustment for comorbidities (Fig. 1B), total costs for stage 2 subjects remained significantly greater than for normal or stage 1 subjects ($4,556 vs. $4,388 and $4,370, P < 0.001), but except for the pharmaceutical component of costs, the difference between stage 1 and normal subjects disappeared.

Table 2 describes the 26,309 patients who were initially assigned to the normal FPG control group and who survived through 2003, displayed by the highest FPG stage observed by 31 December 2003. Most subjects (89.8%) did not progress to a subsequent stage. Nonprogressing patients were younger than those who progressed to either stage 1 or stage 2 (57.5 years vs. 60.2 and 60.7 years, respectively, P < 0.001 for both comparisons), but of similar age to those who progressed to diabetes (57.0 years, NS). Subjects who remained in normal FPG status were much more likely to be women (47.9% vs. 33.3, 34.9, and 39.0%, P < 0.001) and much less likely to have CVD. Subjects who did not progress to a higher FPG stage had significantly lower systolic blood pressure, LDL cholesterol, triglycerides, total cholesterol and BMI observed while in the normal FPG category and significantly higher HDL cholesterol. Nearly one-third (32.1%) of subjects with normal FPG who ultimately progressed to diabetes already had three metabolic syndrome criteria, compared with just 10.5% of subjects whose FPG status remained normal (P < 0.001). Finally, subjects who did not progress incurred significantly lower costs than those who progressed to an IFG stage or to diabetes ($3,785 vs. $4,459, $5,307, and $6,568, P < 0.001 for all comparisons). Inpatient, outpatient, and pharmaceutical components of costs displayed a similar pattern. Adjustment for age, sex, and comorbidities did not affect the comparisons (adjusted data not shown).

Figure 2 redisplays costs by FPG stage, but includes as control subjects only those subjects who survived to the end of 2003 without progressing to a higher FPG stage. In this comparison, age- and sex-adjusted total costs and each component of costs differed significantly between stages (P < 0.001 for all comparisons), being lowest among normal subjects and highest among stage 2 subjects. All these differences remained significant after further adjustment for comorbidities.

Analyzed continuously, each 1 mg/dl of FPG added $25 (P < 0.001) to annual costs when other factors were not accounted for (Table 3). Adjustment for age, sex, and CVD risk factors reduced the independent contribution of FPG to $10 per 1 mg/dl annually (P < 0.001). However, after further adjustment for presence of CVD, the independent contribution of FPG to annual costs was not statistically significantly different from zero.

To our knowledge, our study is the first to estimate medical care costs for patients with subdiabetic abnormal glucose metabolism. Consistent with other research (7), the CVD risk profiles for patients at the new ADA cut point for IFG are not as severe as those for patients at the old cut point. Our results indicate that although costs are greater for patients who meet the 2003 ADA cut point for IFG, they are lower than costs for patients at the higher 1997 cut point. Nonetheless, comparisons to patients with normal FPG suggest that the new cut point identifies a group that has higher costs and greater CVD morbidity than normoglycemic patients. Confounding with comorbidities such as CVD makes the isolation of the effect of elevated glucose difficult (11–13), but our results also indicate that for patients added by the new ADA definition, elevated FPG is not in itself associated with higher medical costs, primarily because these patients are more likely to have CVD. Adjustment for CVD comorbidity (Fig. 1) removes an otherwise statistically significant difference in costs between subjects with normal FPG and those in the 100–109 mg/dl range. CVD-adjusted costs remained significantly higher for subjects with stage 2 IFG, suggesting that at higher levels, abnormal glucose metabolism does add independently to costs. However, when analyzed on a continuous basis, FPG does not independently explain any significant variance in annual costs after accounting for the presence of CVD.

Our analysis comparing those who did and did not progress to a higher FPG state demonstrates that patients with normal FPG are not all equivalent (Table 2). That is, some patients with apparently normal FPG may have a genetic predisposition to abnormal glucose metabolism that has not yet manifested, whereas most others are destined to remain within the normal range. Therefore, cost comparisons between normal subjects and patients currently with IFG that include patients with unmanifested IFG in the normal group, as in our initial analysis, understate the true costs of IFG. Our final analysis (Fig. 2) illustrates this point. After removing patients from the normal group who later progressed to IFG or diabetes, we found substantial cost differences between normal subjects and patients with stage 1 IFG, even after adjustment for comorbidities.

Because of our observational design, assignment of patients to various stages of FPG was vulnerable to ascertainment bias. Although many FPG tests are conducted in the health plan due to routine screening, we suspect that many subjects were tested for clinical reasons we could not observe, particularly in the normal FPG group, in which the mean FPG was >90 mg/dl. Therefore, our cost estimates for control subjects might be overstated. We have attempted to address this limitation by subsequently excluding subjects who later progressed beyond normal FPG.

Although we weighted our cost estimates by months of observation, differential observation time is another potential source of bias if costs were incurred differentially at the end versus the beginning of a given stage. For example, costs for a subject identified in 2002 and followed only to the end of 2003 might differ from those for a subject in the same stage identified in 1998 and followed until progression to a subsequent stage that occurred, say, in 2002. If costs were greater in the latter portion of the observed stage, our estimates probably understate the true cost of IFG. By the same logic, if costs are greatest at the beginning of a stage, as they are for diabetes (5), our estimates would overstate true IFG costs. This latter form of differential follow-up bias might be exacerbated by an interaction with ascertainment bias if an adverse health event that leads to FPG testing is driving early stage costs. Overall, however, average observation times of 4 to 5 years probably minimize the effects of differential follow-up.

In large sample sizes such as ours, even relatively small cost differences can appear statistically significant. For example, in our final analysis (Fig. 2), the age-, sex- and comorbidity-adjusted difference in costs between subjects with stage 1 and stage 2 IFG was a statistically significant $217, an amount that represents about 5% of the total costs. Individually, this cost difference may be small, but it quickly reaches millions when multiplied by the large numbers of people to whom it applies. Our large sample size also minimizes the effect of large outliers on cost analyses. We reanalyzed our data excluding subjects with total costs in excess of 3 SDs beyond the mean (total costs > $31,985) with identical results.

Our multivariate analyses yielded an apparently anomalous result: better LDL cholesterol level and systolic blood pressure appeared to be associated with higher costs. However, further analysis (not shown) revealed that patients with existing CVD had lower values for these measures than those without CVD. For example, patients with a history of myocardial infarction had an average LDL level of 111 mg/dl, compared with 127 mg/dl for patients without a history of myocardial infarction (P < 0.001). Systolic blood pressure followed a similar but less extreme pattern (132 vs. 134 mmHg, P < 0.001). Thus, in our data, lower values for these important risk factors served as markers for costly disease.

We identified a subset of patients with apparently normal FPG who progressed to IFG or diabetes. These patients were substantially costlier while their FPG was normal than their counterparts who did not progress to defined levels of abnormal glucose (IFG or diabetes). Further research should focus on whether costs provide a valid method of identifying persons at greatest risk of progressing to type 2 diabetes.

This study was sponsored by the National Institute of Diabetes, Digestive and Kidney Diseases Grant 1 R21 DK063961-01A1.