Glucose screening and the risk of complications in Type 2 diabetes mellitus☆

1. Introduction 

Type 2 diabetes mellitus (DM2) is estimated to affect 14 million individuals in the United States; approximately one third of these are undiagnosed 1, 2. The duration of DM2 before diagnosis has been estimated to be 7 to 10 years, and the presence of microvascular complications at the time of diagnosis is common 3, 4, 5. The burden of microvascular complications from DM2 is high: diabetes is the leading cause of end-stage renal disease, polyneuropathy, and nontraumatic amputations and is the most common cause of new blindness in adults aged 20 to 74 years [3]. Direct and indirect costs of DM2 in 1997 were estimated to be $98 billion [6]. Results of observational and experimental studies suggest that intensive glycemic control in DM2, once diagnosed, reduces the risk of microvascular complications 7, 8, 9, 10, 11, 12. However, these trials leave open the question of whether early detection of DM2, before symptoms appear, is of net benefit.

Screening asymptomatic individuals for DM2 has considerable prima facie appeal: DM2 is a common and burdensome chronic disease with a long pre-symptomatic period. Although there is effective therapy available, many individuals go undiagnosed for years. Screening has nonetheless been a controversial subject. Selective screening has been recommended by some medical organizations on the basis of various risk factors, including age alone 13, 14. However, these recommendations have not been based on any evidence showing that early detection of DM2 via screening improves outcomes 15, 16. Using Monte Carlo modeling techniques, screening for DM2 was judged to be cost-effective for some population groups, although this model was based only on assumptions about the benefits of screening [17]. Given the lack of evidence linking clinical benefit to screening, there is particular concern regarding the possible harms of screening programs due to false positives and “labeling” effects, and false negatives leading to false reassurance 15, 16.

Although a randomized trial of diabetes screening would, in theory, be the ideal way to measure efficacy, such a trial would be difficult to accomplish because of the large numbers of patients and length of follow-up time required. In the absence of data from randomized trials, case-control methods have been developed to evaluate the effectiveness of screening [18]. Case-control studies have been used to evaluate screening for colorectal, prostate, cervical, and breast cancer 19, 20, 21, 22. To test if there is an effect of screening on DM2 complications, we compared previous receipt of a blood-glucose screening test between subjects who had certain clinically unmistakable microvascular complications of DM2 and a sample of the general study population.

2. Methods 

2.2. Definition of complications 

Complications were defined as symptomatic microvascular end-organ manifestations that were attributable to diabetes. Table 1 shows study criteria for complications. The occurrence of a complication was confirmed through review of physicians' notes indicating such a diagnosis in the medical record. In all instances, complications were considered attributable to DM2 only if this attribution was explicit in the medical record.

Table 1. Criteria for having a symptomatic microvascular complication of DM2

	Category	DM2-related manifestation
	Neurologic	Peripheral neuropathy Autonomic neuropathy Charcot arthropathy
	Ophthalmologic	Laser treatment of retinopathy, macular edema, or retinal detachment Blindness
	Renal	Persistent mean serum creatinine >309 μmol/L (3.5 mg/dL) Chronic dialysis or renal transplant
	Amputations/Ulcer	Non-traumatic lower-extremity amputations Foot ulcers

Abbreviation: DM2, diabetes mellitus type 2.

Operationally, “symptomatic” was taken to mean a disease manifestation that was almost certainly noticeable to patients, or, in other words, that would have some impact on quality of life. Thus, whereas retinopathy in a given patient may have been attributed to diabetes, it was not counted as a complication unless the patient had to undergo laser therapy to treat it or unless the patient was also diagnosed with visual impairment (blindness or partial blindness) as a result of the retinopathy. Microalbuminuria and mild degrees of renal impairment are usually asymptomatic and thus did not count as complications in this study. We chose to define renal complications in this study by conditions that would almost certainly affect patient quality of life—complete renal failure that led to transplant or dialysis or significant chronic renal failure (with mean creatinine levels of >3.5 mg/dL), which almost invariably necessitates changes in diet and other treatments.

Neuropathy complications were all considered symptomatic. Problems such as autonomic neuropathy or Charcot arthropathy would not come to the attention of clinician unless symptomatic. In the case of peripheral neuropathy, some of these patients may be detected by the use of monofilament testing before they present with symptoms; however, once discovered, this diagnosis would lead (at the very least) to noticeable lifestyle changes necessary to prevent injury to the extremities (eg, always wearing shoes, changes in the style of shoe permissible, frequent inspections of the feet, etc.). Foot ulcers and amputations were considered inherently symptomatic; the former, even if painless, requires treatment with antibiotics, dressing changes, restricted weight bearing, etc. Amputations have an obvious impact on quality of life for patients.

2.5. Documentation of glucose tests, potential confounding variables, and enrollment duration 

Reviewers were not blinded to whether individuals were case subjects or control subjects. After eligibility as a study subject was verified, chart reviewers recorded every blood glucose test performed during the 10-year review period (the 10 years before the reference date). For each test, the following information was recorded: date, type of test (fasting or random, oral glucose tolerance test, hemoglobin A1c), result, whether any abnormal result had clinical follow-up, and the clinical intent or indication for the test. This last item was judged by examination of clinicians' notes and was categorized using the following guidelines:

1.Tests for symptoms referable to diabetes. These tests occurred in the setting of classic symptoms of diabetes (e.g. polyuria, polydipsia, or polyphagia) or in the course of investigations of diseases or symptoms where diabetes might be have been a cause or underlying factor (Table 2).

Table 2. Conditions for which diabetes was considered as possible underlying cause

	Condition	Examples
	Ophthalmologic	Blindness/vision loss; retinopathy; macular edema
	Renal	Renal failure; proteinuria
	Neurologic	Numbness, abnormal sensation/pain in extremities
	Amputations	Lower-extremity nontraumatic amputations
	Foot ulcer/infection	Skin breakdown; chronic infection/ulcer in feet
	Atherosclerotic vascular disease	Coronary artery disease/ischemia; angina/chest pain; cerebrovacular disease/ischemia; cerebrovacular accident/stroke; peripheral vascular disease, claudication
	Recurrent/unusual infections	Recurrent vaginal candidiasis; onychomycosis; recurrent rashes or other infections
	Constitutional	Significant weight gain/loss that is reason for visit; fatigue
	Other	Follow-up testing for any previous abnormal glucose; any condition not previously listed, and chart notes indicate diabetes being ruled out

2.Tests without symptoms of diabetes. There were two subtypes of these tests. The first subtype comprised so-called “population screening” tests, where the clinical intent was deliberately to screen for diabetes per se. The second subtype comprised so-called “opportunistic screening” or “case-finding” tests, where the measurement of glucose was incidental to other clinical investigations (e.g. evaluations of acute gastrointestinal illness or follow-up of chronic diseases such as hypertension) and was not driven by concerns about diabetes.

3.Unknown. If after medical record review the clinical intent of the test could not be categorized using the rules enumerated in (1) and (2) or was otherwise ambiguous, this classification was used.

Data were gathered from the medical record on each subject about factors that were possible confounders. These included possible risk factors for diabetic microvascular complications that might also be associated with increased screening activity, such as body mass index (BMI) at or near to the reference date, family history of diabetes (as indicated in the medical record), and number of preventive or health maintenance visits over the review period. The presence or absence in the medical record of three comorbid states (hypertension, coronary artery disease, and hyperlipidemia) during the review period was also recorded. These conditions might also be related to the likelihood of being screened, and, at least in the case of hypertension, may be related to the likelihood of having a microvascular complication.

3. Results 

The most common initial complication type was neurologic, comprising 64% of first complications, followed by ophthalmologic (16.5%), foot ulcer/amputations (13.9%), and renal (5.6%). Compared with control subjects, case subjects were more likely to have hypertension, coronary artery disease, hyperlipidemia, a family history of diabetes, and a higher BMI (Table 3). However, control subjects on average had more health maintenance visits.

Table 3. Characteristics of study subjects

	Variable	Case Subjects (n =303)	Control Subjects (n =303)
	Women, %	52.8	52.8
	Mean age at reference date, y	63.9	63.9
	Family history of diabetes, n (%)	148 (49)	79 (26)
	Hypertensiona, n (%)	206 (68)	116 (38)
	Coronary artery diseasea, n (%)	98 (32)	53 (18)
	Hyperlipidemiaa, n (%)	90 (30)	57 (19)
	Body mass index, mean	31.6	26.9
	Number of preventive visitsb, mean	1.95	2.67

a Comorbid conditions found on chart review to be present before the reference date. b Mean number of visits over 10-year review period.

A total of 4125 glucose tests were recorded. Of these, 3360 (81%) were judged to have occurred without symptoms referable to diabetes and were therefore classified as screening, and 522 tests (13%) were judged to have occurred in the context of symptoms. The clinical intent of 243 tests (6%) could not be judged, and these were categorized as “unknown.” Case subjects had more screening glucose tests than control subjects, with an average of 6.3 tests per case subject over the review period compared with an average of 4.8 tests per control subject. By far the most common form of screening glucose test performed on subjects was a “random” glucose, which denotes that no particular temporal relationship to last caloric intake is required or assumed. Over the 10-year review period, random glucose tests comprised 88% of all screening tests for case subjects and control subjects; fasting glucose tests comprised 10% of tests for case subjects and 7% for control subjects.

Using a definition of screening that was operationalized as “deliberate” screening tests and incidental testing (“case-finding”) in the absence of the suspicion of diabetes, 89% of subjects had at least one screening test over the 10-year review period. Slightly more case subjects than control subjects were screened (91% versus 87%, respectively). Adjusting only for the matching factors, the OR for exposure to a screening test at least once in the review period was 1.7. After adjustment for BMI; number of health maintenance visits; family history of diabetes; and the presence of hypertension, hyperlipidemia, and coronary artery disease during the review period, the adjusted OR for a history of a screening test was 0.87 (95% confidence interval [CI] 0.38–1.98). The number of health maintenance visits over the 10-year review period was independently associated with reduced odds of a complication, with an OR of 0.77 (95% CI 0.67–0.87) (Table 4).

Table 4. Odds ratios for variables from the fully-adjusted model

	Variable	Odds ratio	95% CI
	Glucose screeninga	0.87	0.38–1.98
	Number of health maintenance visits	0.77	0.67–0.87
	Body mass index	1.15	1.10–1.20
	Family history of diabetes	2.31	1.50–3.54
	Hypertension	2.47	1.52–4.00
	Coronary artery disease	1.84	1.05–3.22
	Hyperlipidemia	1.41	0.83–2.41

Abbreviation: CI, confidence interval.

a Unadjusted odds for screening: 1.70; 95% CI 0.98-2.95.

In this population, failure to follow-up an abnormal blood glucose result occurred frequently. For random glucose tests having any degree of abnormality, failure to follow-up occurred 45% of the time for case subjects and 20% of the time for control subjects. Considering only random glucose tests with values >11.1 mmol/L (200 mg/dL), 40% of case subjects compared with 14% of control subjects failed follow-up.

4. Discussion 

The data from this case-control study suggest that persons who had at least one glucose screening event in the 10-year review period had a 13% reduction in the risk of developing severe microvascular complications of DM2 after adjusting for multiple potentially confounding factors. However, the wide CI (0.38–1.98) around the estimated relative risk argues for a cautious interpretation. A 13% reduced risk seems plausible, given that randomized trials have shown that intensive treatment is associated with approximately a 20% decreased risk of complications among persons with DM2 [25].

Misclassification of exposure status is a potential source of bias in this study. If patients presented with symptoms but this was not documented in the medical record, tests may have been misclassified as screening. This would lead to an underestimate of the effectiveness of screening. However, most subjects had undergone many tests categorized as screening, so that all these tests would have had to be misclassified to change the exposure status (because to qualify as having been screened, it need happen only once in 10 years). In addition, the fact that reviewers were not blinded to whether subjects were case subjects or control subjects could potentially lead to bias in the classification of tests as screening/not screening. This was minimized through the use of detailed categorization rules as detailed in the Methods section.

This study focused on more severe endpoints of DM2. This approach was taken based on the rationale that, to be worthwhile, screening should be effective in preventing outcomes that are noticeable to patients (ie, that are in some way symptomatic). Although asymptomatic manifestations are important, their clinical significance is less clear than symptomatic manifestations because of uncertainty over whether or how rapidly the former progresses to affect how patients feel and function. In addition, we preferred the use of “symptomatic” complications to define our cases because ascertainment of these clinically evident complications is more certain as patients would likely seek care for such complications, and this would be reflected in the medical records reviewed. In contrast, ascertainment of early, asymptomatic complications is less certain, and in the absence of progression, these complications would have minimal impact on patient quality of life.

It is possible that differences in provider type might have introduced confounding. For example, “good” physicians might have aggressively screened for diabetes and when it was discovered might have managed the disease in an aggressive manner that yielded better outcomes. However, we did not record data that identify the primary physician of a given patient. Nonetheless, Group Health Cooperative is a large organization, with 500,000 enrollees. Given our random selection of patients and the large panel of clinicians providing primary care, it is likely that our subjects were cared for by a large sample of Group Health Cooperative physicians, and, thus, the impact of a few standout clinicians would not be substantial.

Although we were able to adjust for the presence of hypertension, hyperlipidemia, and coronary artery disease as potentially confounding factors, we did not have data to measure severity of these conditions. To the extent that these conditions may contribute to diabetic complications (particularly hypertension) in proportion to severity, residual confounding may exist.

Had there been a higher degree of response to an abnormal screening test, it is likely the efficacy of testing would have been greater than that observed here. For screening to be effective, abnormal tests must be followed up. Moreover, there was a substantial difference between case subjects and control subjects in terms of follow-up patterns. The reasons for this difference are not clear, nor is it clear whether this failure was primarily that of the physician or of the patient. Case subjects had fewer health maintenance visits than control subjects (Table 3), perhaps suggesting a difference between case subjects and control subjects in terms of health consciousness.

We did not have socioeconomic status data for our study subjects and thus did not control for difference between case subjects and control subjects. The majority of enrollees in the Group Health Cooperative HMO are such by virtue of employment benefits. However, co-payments and other out-of-pocket costs have been modest for enrollees; thus, for those few indigent patients or the working poor, financial barriers to care are not substantial. It is unlikely that adjusting for socioeconomic status would have significantly changed our results.

That the number of health maintenance visits was independently associated with lower risk of complications is of some interest. Other than screening for diabetes that may arise in this context, counseling about diet and exercise may also theoretically lead to reduced risk of diabetic complications. A regular pattern of health maintenance visits may be a marker for aspects of lifestyle or health consciousness, which may lead to reduced risk of diabetes, reduced delay of diagnosis, or more careful management of diabetes once diagnosed, thus having an impact on the risk of complications. This study was unable to measure or evaluate in any greater detail this complex phenomenon, which can be only partially addressed via a simple recording of the number of health maintenance visits.

Although it is clear that macrovascular complications such as stroke and myocardial infarction are responsible for the vast majority of excess mortality of DM2 [3], there continues to be no firm evidence that tight glycemic control reduces the risk of macrovascular endpoints [25]. It may be true that through DM2 screening we would identify individuals who would subsequently be found to have risk factors for macrovascular disease (eg, hypertension and hyperlipidemia). Subsequent treatment of these disorders could then reduce their risk of macrovascular disease. However, our goal was to focus on diabetic complications that could be directly affected by the treatment of diabetes itself. It was for this reason that macrovascular endpoints were not counted as complications in this study. If in the future glycemic control were linked with macrovascular complications, then this would indicate the need to evaluate the effectiveness of screening in preventing those outcomes. Future studies of the effectiveness of screening for diabetes might focus on measuring the benefits that ensue from identifying, and treating, macrovascular disease risk factors.

In conclusion, these findings suggest that diabetes screening may be associated with a lower risk of the development of certain diabetic complications. However, the incomplete follow-up of many persons screened as positive, along with the wide confidence limits surrounding the estimate of effectiveness, argue for a cautious interpretation. By itself, this study does not support a policy of screening asymptomatic adults. However, these data do suggest the need for further studies to address the effectiveness of screening in reducing microvascular diabetic complications, perhaps by repeating this study in other populations and aggregating data to arrive at a pooled estimate of effectiveness.