Impact of time since last meal on the gestational glucose challenge test

Impact of time since last meal on the gestational glucose challenge test

Sermer et al. Volume 171, Number 3 Am J Obstet Gynecol 24. Flower R]. Drugs which inhibit prostaglandin biosynthesis. Pharmacol Rev 1974;26:33-67. 2...

991KB Sizes 166 Downloads 36 Views

Sermer et al.

Volume 171, Number 3 Am J Obstet Gynecol

24. Flower R]. Drugs which inhibit prostaglandin biosynthesis. Pharmacol Rev 1974;26:33-67. 25. Baylis C. Renal effects of cyclooxygenase inhibition in the pregnant rat. Am] Physiol 1987;253:FI58-63. 26. Khraibi AA, Heublein DM, Burnett ]C, Walker KR, Knox FG. Renal interstitial hydrostatic pressure and ANF in exaggerated natriuresis of the SHR. Am] Physiol 1990; 258:RI380-5.

27. Keller R. Atrial natriuretic factor has a direct, prostaglandin-independent action on kidneys. Can] Physiol Pharmacal 1982;60:1078-82. 28. Weiner C, Zhu LK, Thompson L, Herrig ], Chestnut D. Effect of pregnancy on endothelium and smooth muscle: their role in reduced adrenergic sensitivity. Am] Physiol 1991;261:H 1275-83.

Impact of time since last meal on the gestational glucose challenge test The Toronto Tri-Hospital Gestational Diabetes Project Mathew Senner, MD, C. David Naylor, MD, DPhil, Douglas J. Gare, MD, Anne B. Kenshole, MD, J.W.K. Ritchie, MD, Dan Farine, MD, Howard R. Cohen, MD, Karen McArthur, MD, Stephen Holzapfel, MD, Anne Biringer, MD, Erluo Chen, MB, MPH, Kenneth I. Cadesky, MD, Ellen M. Greenblatt, MD, Nicholas A. Leyland, MD, Heather S. Morris, MD, Jeff A. Bloom, MD, and Yoel B. Abells, MD, for the Toronto Tri-Hospital Gestational Diabetes Investigators Toronto, Ontario, Canada OBJECTIVE: The purpose of the study was to evaluate the impact of time since the last meal on the glucose challenge test and to find cut points that are most likely to predict the outcome of the oral glucose tolerance test in patients screened for gestational diabetes. STUDY DESIGN: This prospective analytic cohort study was carried out at the University of Toronto Perinatal Complex. A 50 gm glucose load was given at 26 weeks' gestation and the time since previous meal ingestion was recorded. At 28 weeks' gestation a 100 gm oral glucose tolerance test was administered. A total of 4274 eligible patients were screened. RESULTS: Time since the last meal had a marked effect on mean plasma glucose. Receiver-operator characteristic curve analysis with National Diabetes Data Group criteria to interpret the oral glucose tolerance allowed the selection of the most efficient cut pOints for the glucose challenge test on the basis of time since the last meal. These cut points were 8.2, 7.9, and 8.3 mmol/L for elapsed postprandial times of < 2, 2 to 3, and > 3 hours, respectively. With this change from the current threshold of 7.8 mmol/L the number of patients with a positive screening test dropped from 18.5% to 13.7%. There was an increase in positive predictive value from 14.4% to 18.7%. The rate of patient misciassification fell from 18.0% to 13.1 %. CONCLUSION: We suggest that screening strategies for detection of gestational diabetes be reconsidered, to account for the impact of variable postprandial status on the test results. (AM J OBSTET GVNECOL 1994;171:607-16.)

Key words: Gestational diabetes, glucose challenge test, screening

From the University of Toronto Perinatal Complex. Supported by Ontario Ministry of Health grant No. 02650. Received for publication July 6, 1993; revised January 4, 1994; accepted March 15, 1994. Reprint requests: M. Sermer, MD, 6EN-223, The Toronto Hospital, Toronto, Ontario, Canada M5G 2C7. Copyright © 1994 by Moslry-Year Book, Inc. 0002-9378/94 $3.00 + 0 6/1/56210

Because of the potential impact of gestational diabetes mellitus on maternal and fetal morbidity, j·9 the Second (1984) and Third International Workshop Conferences on Gestational Diabetes and the American Diabetes Association have recommended routine screening of all pregnancies for gestational diabetes

607

608

Sermer et al.

mellitus. I. 10 The oral glucose tolerance test (GIT) remains the reference test for establishing the diagnosis of gestational diabetes mellitus, but it is impractical as a universal screening procedure. Hence the issue is how and whom to screen. Urine testing for glucosuria continues to be performed by a majority of practitioners, but physiologic changes in pregnancy mean that up to 50% of pregnant patients will have glucosuria on urinalysis and the predictive value of urinalysis is accordingly low. II Measurement of glycosylated hemoglobin is an insensitive maneuver. I2 -I6 The random plasma glucose has been advocated for its ease of administration and low cost. 17-19 No waiting time is required, and a standard glucose load does not have to be ingested. However, the potential accuracy of the random plasma glucose is inherently limited by the lack of standardization of timing from the previous meal and the attendant lack of standardization of the carbohydrate load. The Second International Workshop Conference on Gestational Diabetes accordingly recommended that all pregnant women undergo a I-hour, 50 gm glucose challenge test between 24 and 28 weeks of gestation, using a threshold value of 140 mg/dl (7.8 mmoVL), without regard to the time. or the nature of the previous meal. I The same verdict was upheld by the Third International Workshop several years later. The glucose challenge test was first proposed by O'Sullivan et al. 20 in 1973 after its sensitivity and specificity for predicting an abnormal oral GIT were shown to be 79% and 87%, respectively. In a retrospective analysis of the data from a group of 752 women who underwent both glucose challenge test and oral GIT in 1956 to 1957 the glucose challenge test threshold level was established at 143 mg/dl and later rounded down to 140 mg/dl for ease of recollection. Of the 752 women, 361 were ~25 years old, and 16 of 19 women with gestational diabetes mellitus fell into this subgroup. O'Sullivan 21 and O'Sullivan et al. 20 accordingly recommended that it would be most efficient to confine the glucose challenge test to women ~ 25 years old. Since then researchers have not agreed on the threshold values for the glucose challenge test, the range of suggested values being between 130 to 150 mg/dl (7.2 to 8.3 mmoVL).I. 21-27 There is also a controversy about the impact of testing patients in the fasting or fed state 25 and the role, if any, of selective screening on the basis of classic risk factors for gestational diabetes mellitus. 26. 27 Uncertainty is heightened by the fact that no study since that of O'Sullivan has administered both the glucose challenge test and oral GIT to subjects without regard for the glucose challenge test results. In consequence, proposals to shift the threshold values for the glucose challenge test cannot be calibrated against the true misclassification rate.

September 1994 Am J Obstet Gynecol

These and other persisting controversies in gestational diabetes led us to initiate the Toronto TriHospital Gestational Diabetes Project, a prospective analytic cohort study addressing a range of issues dealing with carbohydrate intolerance during pregnancy. A pertinent design feature of the project was the recruitment of subjects with the understanding that they will be proceeding to oral GIT, regardless of glucose challenge test results. The goal of this report from the project is to establish the impact of time since the previous meal on glucose challenge test and to find glucose challenge test threshold values that are most likely to predict normal or abnormal oral GIT, as defined by the National Diabetes Data Group. We also present threshold values for the oral GIT when interpreted according to the criteria of Carpenter and Coustan!B. 29 Methods

Three teaching hospitals forming the University of Toronto Perinatal Complex took part in this prospective analytic cohort study between September 1989 and March 1992. This project had a prior approval by the ethics committees of the University of Toronto and the three participating teaching hospitals (Toronto Hospital, Mount Sinai Hospital, and Women's College Hospital). Our aim was to design and implement a large, simple study that would address a variety of issues with adequate statistical power. All pregnant women ~ 24 years old at delivery were eligible for participation in the study, provided that there was no history of preexisting diabetes and they were seen by their physician before 24 weeks' gestation. To remain in the study, patients had to be delivered after the twentieth week of gestation. At 26 weeks' gestation (± 7 days) consenting eligible patients underwent a glucose challenge test. A 50 gm glucose load was given, followed by a plasma blood glucose test 1 hour later. Just before glucose challenge test administration, time since the previous meal, demographic data, and history of classic risk factors were recorded. For simplicity "last meal" meant last ingestion of any food or fluid. We assumed that there would be imprecision in reportage in practice and that the effect of this imprecision in the study would be to weaken any statistical relationship between time since the last meal and glucose challenge test results. Hence any effect demonstrable here would err on the side of underestimation and be hugely relevant to practice. All patients were then asked to return at 28 weeks' gestation (±7 days) for a 100 gm oral-GIT. The oral GIT was performed in the morning after an overnight fast of at least 8 hours but no more than 14 hours and after 3 days of a 150 gm carbohydrate diet and unre-

Sermer et al.

Volume 171, Number 3 Am J Obstet Gynecol

609

Table I. Oral CIT cutoff values for National Diabetes Data Croup criteria and for criteria of Carpenter and Coustan 28 Plasma glucose value National Diabetes Data Group criteria mgldl

Fasting

1 hr 2 hr 3 hr

I

105 190 165 145

CarpenterlCoustan criteria

mmollL

mgldl

5.8 10.5 9.1 8.0

95 180 155 140

I

mmollL

5.3 10.0 8.6 7.8

Adapted from O'Sullivan et al!°

stricted physical activity. Clucodex (Rougier) was used for the 50 and 100 gm glucose load. The glucose oxidase method was used for plasma blood glucose determination. Plasma glucose analysis was performed only in one institution, the Toronto Hospital, to avoid bias or imprecision stemming from technical differences in plasma glucose determination present in different laboratories. For the primary analysis the oral CIT was considered positive if at least two of the four blood sugar values were greater than or equal to the cutoff points as established by the National Diabetes Data Croup 30 (Table I). A secondary analysis with the criteria of Carpenter and Coustan 28 was also performed given recent outcome evidence in support of these criteria29 and laboratory evidence 3l suggesting use of an inappropriate conversion algorithm when the original criteria of O'Sullivan et al!O were updated by the National Diabetes Data Croup 30 for use with modern glucose oxidase methods. This analysis, however, was not prespecified. The target sample size of 4000 patients was chosen on the basis of an explicit trade-off between augmented precision from a larger sample size and the feasibility of the study. In particular, the marginal reduction in the 95% confidence interval around the anticipated sensitivity and specificity of the glucose challenge test declined rapidly from 4000 subjects upward. With this sample size the 95% confidence interval was ± 6% around the anticipated sensitivity of 80% for the glucose challenge test when measured against the diagnosis of gestational diabetes mellitus on the full 100 gm oral CIT. The 95% confidence interval was ± 1% around the anticipated specificity of 90%. The anticipated sensitivity and specificity were selected on the basis of the original findings of O'Sullivan et al. 20 For analysis of the data patients were separated into groups determined by the time interval between the last meal and glucose challenge test administration. The groups were < 1 hour, 1 to 2 hours, 2 to 3 hours, 3 to 4 hours, and > 4 hours; the latter two groups were

collapsed for some analyses, to avoid sparse data. Comparisons of baseline characteristics among the groups were made with X2 or Fisher's exact test for categoric variables and with analysis of variance for continuous variables. To control for multiple comparisons of baseline variables, a Bonferroni correction was applied to reduce the threshold for nominal statistical significance to 0.004 (derived from 0.05: 11 because 11 characteristics were compared). The overall comparison for differences in glucose challenge test on the basis of time since last meal was made with analysis of variance, and specific between-group pairwise comparisons were carried out with Scheffe's test to hold the a or type I error rate constant at 0.05 in spite of multiple comparisons. Receiver-operator characteristic curve analysis was performed for each subgroup by time since the last meal, to determine the plasma glucose threshold value from the glucose challenge test that yielded the highest combined sensitivity and specificity, where gestational diabetes mellitus on the 100 gm oral CIT was the outcome. 32 Subgroup comparisons of area under the receiver-operator characteristic curves were carried out for the conventional threshold of 7.8 mmol/L versus the optimized threshold, by means of published methods." To compare the misclassification rates with the single conventional glucose challenge test threshold (7.8 mmollL) with the cumulative misclassification rates with thresholds adjusted for time since the last meal across all subgroups, changes in sensitivity and specificity were weighted by the proportions of patients with positive and negative oral CITs, respectively. The null hypothesis of the equivalence of weighted proportions was tested with a weighted two-sample test of binomial proportions, calibrated against the Z-normal distribution. Last, although receiver-operator characteristic curve analysis presupposes optimization on the basis of combined sensitivity and specificity, we explored two alternative optimization thresholds based on arbitrary levels of sensitivity alone. The thresholds were chosen as follows. All sensitivity values for 0.1 mmol changes in

610

Sermer et al.

September 1994 Am J Obstet Gynecol

Table II. Comparisons for demographic information and medical history on basis of time since previous meal for 3836 women completing both glucose challenge test and oral CIT Time since last meal Variables

Age (yr, mean and SD) Body mass index (kg/m", mean and SD) Race White Black Oriental Other Gravidity (mean and SD) Parity (mean and SD) Family history of diabetes Previous pregnancy complicated by Gestational diabetes mellitus Macrosomia Stillbirth Anomalies Preeclampsia

(n

<1 hr = 1109)

I

(n

1-2 hr = 1200)

I

(n

2-3 hr = 773)

I

(n

>3 hr = 754)

31.3 (4.2) 38.9 (6.2)

31.3 (4.2) 39.2 (7.4)

31.3 (4.3) 39.0 (7.1)

30.9 (4.2) 39.6 (7.8)

799 53 74 33 2.0 0.6 187

820 46 85 40 2.1 0.6 177

523 28 47 26 2.1 0.6 115

451 41 57 31 2.1 0.7 107

(83.3%) (5.5%) (7.7%) (3.5%) (1.2) (0.7) (17.0%)

(82.7%) (4.6%) (8.6%) (4.1%) (1.2) (0.8) (14.8%)

(83.8%) (4.5%) (7.5%) (4.2%) (1.3) (0.8) (14.9%)

(77.8%) (7.1%) (9.8%) (5.3%) (1.2) (0.9) (14.3%)

Significance

P = 0.08 P = 0.19 p = 0.18

P = 0.38 P = 0.06 P = 0.34

17 (2.6%)

22 (3.1%)

13 (2.8%)

19 (4.2%)

P = 0.49

68 7 23 31

68 11 42 42

46 18 24 28

37 11 24 24

P = 0.63 P = 0.01 P = 0.24 P = 0.79

(10.5%) (1.1%) (3.6%) (4.5%)

(9.6%) (1.6%) (5.9%) (5.9%)

(9.8%) (3.8%) (5.1%) (5.9%)

(8.2%) (2.4%) (5.3%) (5.3%)

No data on race were collected in the single-center run-in phase, hence the discrepancy in total subjects in the "race" section. Pregnancy history was only applicable for 1505 women with more than one pregnancy; 99.1% of data were available for analysis.

the glucose challenge test were examined overall and by time since last meal for both sets of oral CIT interpretive criteria. The sensitivity nearest to 80% and 90% was chosen, and related glucose challenge test cut points along with other key test characteristics were recorded. Where the sensitivity was constant across two or three increments of 0.1 mmol on the glucose challenge test, the threshold with the highest specificity was used. All analyses were performed in SAS (SAS Institute, Cary, N.C.). Results

During the study period 75 participating physicians delivered 14,007 eligible patients, of whom 4274 (31%) participated in the study. The balance of patients either refused to participate in the trial or were not approached by the participating physicians. Mter the initial glucose challenge test screen, 3836 (90%) continued with the diagnostic oral CIT. The prevalence of gestational diabetes mellitus in this population was 3.8%. Of the 438 who did not proceed with the oral CIT, 19 had a positive glucose challenge test (i.e., exceeded the 7.8 mmol!L threshold value). The proportion with positive glucose challenge tests was higher among those who proceeded to the oral CIT (769/3836 vs 19/438, P < 0.001), indicative of self-selection among noncompliers. Patient characteristics. The demographic data and obstetric histories, including classic risk factors for de-

velopment of gestational diabetes mellitus, were similar across subjects grouped by time since the last meal. Statistical assessments showed that none of the overall differences were larger than would be compatible with the play of chance alone (Table II). Variation by time since last meal. Mean plasma glucose on the glucose challenge test demonstrated a J-shaped relationship, first declining as the time since last meal increased and then rising again starting in the 3- to 4-hour time window (Fig. 1). The overall time effect was highly significant (p < 0.0001 with analysis of variance). Individual pairwise comparisons showed significant differences for the following pairs: > 4 hours versus < 1, 1 to 2, and 2 to 3 hours and 3 to 4 hours versus 1 to 2 hours and 2 to 3 hours. Failure of some pairwise comparisons to attain nominal significance is expected given the J-shaped relationship between time since last meal and glucose challenge test values; moreover, the application of Scheffe's test to correct for multiple pairwise comparisons is conservative. Simple examination of time-specific means and SEs in Table III illustrates that other pairwise comparisons bordered on attaining statistical significance. Time-specific cut points. Receiver-operator characteristic curve analysis for the National Diabetes Data Croup criteria allowed the selection 'of the most efficient cut points for glucose challenge test on the basis of time since the previous meal. The optimal combination of sensitivity and specificity occurred at the follow-

Sermer et al.

Volume 171, Number 3 Am J Obstet Gynecol

611

7

6.89

:::r ;::;. 0

E E

6.8

CD II)

0

u

= (;

6.6

ftII

E II) ftII

a::

6.4

6.2

<1

1·2 2·3 3·4 :t:ime Since the Last Meal (hours)

>4

Fig. 1. Mean glucose challenge test plasma glucose plotted for each I-hour interval since previous meal.

Table III. Mean glucose challenge test plasma glucose value on basis of time since previous meal Time (hr)

No.

Mean plasma glucose value (mmolIL)

SD

SE

<1 1-2 2-3 3-4 >4

1235 1340 853 338 488

6.474 6.294 6.367 6.749 6.887

1.587 1.480 1.433 1.479 1.570

0.045 0.040 0.049 0.080 0.071

99.5% of data available for analysis (4254/4274).

ing cut points: < 2 hours, 8.2 mmo1!L; 2 to 3 hours, 7.9 mmo1!L; and > 3 hours, 8.3 mmo1!L. Comparisons between these values and the currently recommended 7.8 mmo1!L cutoff value are summarized in Table N. The comparisons are made with the prevalence of gestational diabetes mellitus adjusted to a uniform 3.8% across all time categories. Results were identical for cut points, with minor differences in test performance parameters when calculations were repeated with the small variations in prevalence that occurred among subgroups by time since the last meal. The threshold values derived from the receiver-operator characteristic curve analysis resulted in an overall decrease in sensitivity from 76.6% to 73.8%, an increase in specificity from 82.2% to 87.4%, an increase in positive predictive value from 14.4% to 18.7%, a small change in negative predictive value from 98.9% to 98.8%, and a decrease in misclassified patients from 18.0% to 13.1%. For any given time frame, compared within its own subgroup, the new threshold value did not lead to a statistically significant improvement in predictive performance

compared with the benchmark value of 7.8 mmo1!L. However, with the weighted two-sample test of binomial proportions to assess the significance of the overall change in misclassification rates, the resultant t statistic was 13.83, p < 0.0001. There were a total of 4256 patients who had a glucose challenge test result and the time since the previous meal documented. With the new threshold values the number of patients with a positive glucose challenge test decreased from 788 to 583. Thus the proportion of positive screening tests dropped from 18.5% to 13.7%. If only fully documented patients are included (i.e., those who also had an oral GTT), then the number of patients with a positive glucose challenge test decreased from 769 to 572. There were 3836 fully documented patients, and the percentage of patients with a positive screening test dropped in this group from 20.0% to 14.9%. The overall decrease in misclassification was obviously driven largely by the substantial changes in the false-positive rate with nominal increases in the false-negative rates. A parallel analysis was performed

612 Sermer et al.

September 1994 Am J Obstet Gynecol

Table IV. Comparison between 7.8 mmoVL glucose challenge test threshold value and threshold values generated by receiver-operator characteristic curve analysis, with National Diabetes Data Group criteria as reference standard Last meal <1 hr

1-2 hr

2-3 hr

All patients

>3 hr

Cut point Cut point Cut point Cut point Cut point Cut point Cut point Cut point 7.8 8.2 7.8 8.2 7.8 7.9 7.8 8.3 Cut point New cut mmollL mmol!L mmol!L mmol!L mmol!L mmol!L mmollL mmol!L 7.8 mmollL points Patient No. Area under receiveroperator characteristic curve Sensitivity Specificity Positive predictive value Negative predictive value Prevalence Likelihood ratio of positive test Likelihood ratio of negative test PERERRI PERERR2 PERERR

1109 0.82

1109 0.84

1200 0.76

1200 0.78

773 0.79

773 0.80

754 0.82

754 0.83

3836 0.80

3836 0.81

81.6 82.7 15.7

78.9 88.1 20.8

65.8 85.6 15.2

65.8 91.0 22.3

73.7 84.5 15.8

73.7 86.5 17.7

89.5 73.6 11.8

84.2 81.6 15.3

76.6 82.2 14.4

73.8 87.4 18.7

99.1

99.1

98.4

98.5

98.8

98.8

99.4

99.2

98.9

98.8

3.8 4.73

3.8 6.66

3.8 4.55

3.8 7.27

3.8 4.76

3.8 5.45

3.8 3.39

3.8 4.58

3.8 4.42

3.8 5.88

0.22

0.24

0.40

0.38

0.31

0.30

0.14

0.19

0.28

0.30

16.6 0.7 17.3

11.4 0.8 12.2

13.9 1.3 15.2

8.7 1.3 10.0

14.9 1.0 15.9

13.0 1.0 14.0

25.4 0.4 25.8

17.7 0.6 18.3

17.1 0.9 18.0

12.1 1.0 13.1

Sensitivity, specificity, positive predictive value, negative predictive value, and prevalence are expressed as percentages. PERERR1, Percentage of misclassification as false positive = fp/total; PERERR2, percentage of misclassification as false negative = fnltotal; PERERR, overall percentage of misclassifications = (fp + fn)/total. Conversion: 7.8 mmoVL = 140 mgldl, 7.9 mmoVL == 142 mgldl, 8.2 mmoVL = 148 mg/dl, 8.3 mmoVL = 150 mgldl.

with the criteria of Carpenter and Coustan. 28 • 29 As shown in Table V, use of 7.8 mmoVL threshold on the glucose challenge test is accompanied by a clear drop in sensitivity and a concomitant rise in the false-negative rate when those criteria are applied. There are offsetting increases in specificity so that the overall error rate is comparable to that found when the National Diabetes Data Group criteria are used. It is assumed, however, that those who favor a more liberal diagnosis of gestational diabetes, as occurs with the Carpenter and Coustan criteria, are also less likely to accept the reduced sensitivity that occurs with a glucose challenge test threshold of 7.8 mmoVL. The receiver-operator characteristic curve analysis for subgroups by time since last meal (also shown in Table V) allowed the selection of cut points with optimal combination of sensitivity and specificity. These cut points were < I hour, 7.3 mmoVL; I to 2 hours, 7.1 mmoVL; 2 to 3 hours, 6.9 mmoVL; and > 3 hours, 7.6 mmoVL. The receiver-operator characteristic curve analysis redresses the above mentioned loss of sensitivity. In so doing, the overall error rate is driven up dramatically. The marginal efficiency of the glucose challenge test as a screening test apparently falls sharply as the oral GTT criteria are liberalized, owing to the large rise in the glucose challenge test

false-positive rate. Results without adjusting for time since last meal are similar, with the glucose challenge test curve optimized overall cut point of 7.3 mmoVL, sensitivity of 80.4%, specificity of 74.6%, a 23.6% falsepositive rate, and 1.4% false-negative rate. Fixed sensitivity thresholds. As noted above, receiver-operator characteristic curve analysis determines the threshold with the maximum combined sensitivity and specificity. For practitioners who are less concerned about false-positive glucose challenge test results and increased numbers of oral GTTs, the cornerstone of threshold setting may be attainment of a certain minimum sensitivity. Table VI shows the glucose challenge test threshold values and related test characteristics parameters for sensitivities of ~ 80% and ~ 90%. These are reported overall (for practitioners who choose not to inquire about time since last meal) and by time since last meal subgroups. The values are also shown for each set of oral GTT criteria (National Diabetes Data Group and Carpenter and Coustan). As would be expected, higher sensitivity leads to a lower specificity (i.e., a higher proportion of false-positive glucose challenge test results). With the liberalized oral GTT criteria of Carpenter and Coustan the false-positive rates are uniformly and dramatically higher compared with glucose

Sermer et al.

Volume 171, Number 3 Am J Obstet Gynecol

613

Table v. Comparison between 7.8 mmol/L glucose challenge test threshold value and threshold values generated by receiver-operator characteristic curve analysis, with criteria of Carpenter and Coustan as reference Last meal

1-2 hr

<1 hr

>3 hr

All patients

Cut point

Cut point

Cut point

Cut point

Cut point

Cut point

Cut point

Cut point

mmol/L

mmol/L

mmol/L

mmol/L

mmol/L

mmol/L

mmol/L

mmol/L

7.6

Cut point 7.8 mmol/L

New cut points

1109 0.75

1109 0.79

1200 0.76

1200 0.80

773 0.74

773 0.81

754 0.76

754 0.79

3836 0.76

3836 0.80

65.8 84.4 23.8

82.3 75.0 19.6

65.8 87.0 27.3

83.5 75.6 20.2

62.3 85.3 23.9

96.2 66.0 17.3

77.8 74.7 18.6

87.0 71.4 18.4

67.5 83.5 23.3

86.4 72.7 19.1

97.1

98.3

97.2

98.4

96.8

99.6

97.8

98.7

97.2

98.6

7.8

Patient No. Area under receiver-operator characteristic curve Sensitivity Specificity Positive predictive value Negative predictive value Prevalence Likelihood ratio of positive test Likelihood ratio of negative test PERERRI PERERR2 PERERR

2-3 hr

7.3

7.8

7.1

7.8

6.9

7.8

6.9 4.22

6.9 3.29

6.9 5.06

6.9 3.42

6.9 4.24

6.9 2.83

6.9 3.08

6.9 3.04

6.9 4.09

6.9 3.16

0.41

0.24

0.39

0.22

0.44

0.06

0.30

0.18

0.39

0.19

14.5 2.4 16.9

23.3 1.2 24.5

12.1 2.4 14.5

22.7 1.1 23.8

13.7 2.6 16.3

31.7 0.3 32.0

23.6 1.5 25.1

26.6 0.9 27.5

15.4 2.2 17.6

25.5 0.9 26.4

Sensitivity, specificity, positive predictive value, negative predictive value, and prevalence are expressed as percentages. PERERR1, Percentage of misclassification as false positive = ip/total; PERERR2, percentage of misclassification as false negative = fnltotal; PERERR, overall percentage of misclassifications = (fp + fn)/total. Conversion: 6.9 mmoVL = 124 mg/dl, 7.1 mmoVL = 128 mg/dl, 7.3 mmoVL = 132 mg/dl, 7.6 mmoVL = 137 mg/dl, 7.8 mmoVL = 140 mg/dl.

challenge test performance at equivalent sensitivity for the National Diabetes Data Group criteria.

Comment Intuitively it is reasonable to assume that timing of a meal before the glucose challenge test will have an impact on plasma glucose and eventually on the efficiency of the glucose challenge test as a screening test. This large, prospective study demonstrates that there is a significant difference in mean glucose challenge test plasma glucose values, depending on the timing of the previous meal. Demonstrated similarities between the groups make it unlikely that the differences in plasma glucose that were seen between the groups are because of confounding variables. In our view these postprandial changes in the mean glucose challenge test results suggest the need to establish new blood glucose threshold values if current policies of administering the test without regard to the timing of the last meal are to be maintained. Receiver-operator characteristic curve analysis allowed selection of the most efficient cutoff values for prediction of a normal or abnormal oral GTT results. For any given time frame, compared within its own

subgroup, the new threshold value did not lead to a statistically significant improvement in predictive performance compared with the benchmark value of 7.8 mmol/L. However, just as with O'Sullivan's original choice of7.8 mmol/L as a threshold, the new thresholds do represent the point of maximum area under the receiver-operator characteristic curve, with steady diminution in area moving up or down from the new threshold within a given postprandial time grouping. Lack of statistical difference is also partly explicable by the low prevalence of gestational diabetes mellitus and particular attenuation of subgroup-specific statistical power. Thus when the impact of the adjusted thresholds is cumulated across all time categories since the last meal, the new threshold values led to a highly significant 4.9% decrease in overall misclassification of patients (from 18.0% to 13.1%). Positive predictive value improved by 4.3% (from 14.4% to 18.7%), and specificity improved by 5.2% (from 82.2% to 87.4%). These gains are partially offset by a small loss in sensitivity (from 76.6% to 73.8%). The adoption of these new threshold values would be responsible for improved efficiency of the glucose challenge test and a decreased frequency with which the

614

Sermer et al.

September 1994 Am J Obstet Gynecol

Table VI. Alternative threshold based on ensuring high sensitivity National Diabetes Data Group criteria Last meal

<1 hr

Sensitivity Specificity PERERRI PERERR2

2-3 hr

1-2 hr

>3 hr

Overall

Cut point ? 7.8 mmol/L

Cut point ? 7.2 mmol/L

Cut point ? 7.1 mmol/L

Cut point ?6.8 mmol/L

Cut point ? 7.5 mmol/L

Cut point ? 7.0 mmol/L

Cut point ?8.4 mmol/L

Cut point

?8.0 mmol/L

Cut point ? 7.5 mmol/L

Cut point ? 7.0 mmol/L

8.18 82.7 16.6 0.7

90.9 71.5 27.4 0.3

81.3 74.0 25.0 0.7

89.6 67.4 31.4 0.4

81.1 78.8 20.4 0.7

87.9 66.1 32.6 0.5

77.4 83.1 16.3 0.9

90.3 76.6 22.5 0.4

80.7 76.7 22.4 0.7

89.0 66.2 32.5 0.4

Overall: In this column time since the previous meal was not taken into consideration. lbe cut point in each category was determined by the sensitivity that was closest to 80% and 90%, respectively. Sensitivity and specificity are expressed as percentages. PERERR1, Percentage of misclassification as false positive = fp/total; PERERR2, percentage of misclassification as false negative = fnltotal. Conversion: 6.6 mmol!L = 119 mg/dl, 6.8 mmol!L = 123 mg/dl, 6.9 mmol!L = 124 mg/dl, 7.0 mmol!L = 126 mg/dl, 7.1 mmol!L = 128 mg/dl, 7.2 mmol!L = 130 mg/dl, 7.3 mmol!L = 132 mg/dl, 7.5 mmol!L = 135 mg/dl, 7.8 mmol!L = 140 mg/dl, 8.0 mmol!L = 144 mg/dl, 8.4 mmol!L = 151 mg/dl.

diagnostic and potentially unpleasant oral CIT would have to be administered. In particular, these threshold values decreased the proportion of positive glucose challenge tests from 18.5% to 13.7% (or from 20.0% to 14.9% if only the subgroup that went on to oral CIT in this study is considered). This apparently small gain multiplies out to a substantial logistic and cost saving given the large numbers of births in North America each year. For example, there are approximately 4 million births across Canada and the United States, jurisdictions where there is continual pressure for universal glucose challenge test screening of all expectant women. Assuming our results are generalizable to women < 24 years old and assuming that universal screening were to be adopted, the use of time-specific criteria for defining a positive glucose challenge test would lead to 200,000 fewer and oral CITs per year. Thus because of the large number of patients tested even a relatively small improvement in the accuracy of the screening test results in a meaningful decrease in the total number of patients requiring oral CIT. Moreover, the maneuver is simple, entailing one question to the patient, and carries no risks, costs, or inconvenience. There will obviously be concern that these revised thresholds will lead to some missed cases of gestational diabetes. However, using the unadjusted 7.8 mmol!L threshold instead of the proposed thresholds means that 50 more oral CITs are performed to avoid missing one case of gestational diabetes by the National Diabetes Data Croup criteria. Similar efficiency considerations bear notice when the screening performance of the glucose challenge test with respect to the oral CIT interpreted according to

the Carpenter and Coustan criteria is examined. Overall error rates are similar for a glucose challenge test threshold of 7.8 mmol!L with either set of criteria. However, the more liberal Carpenter and Coustan criteria combined with the conventional glucose challenge test threshold (7.8 mmol!L) inevitably result in increases in the number of cases missed (i.e., a higher false-negative rate). If receiver-operator characteristic curve analysis is used on the overall data set, it emerges that the best overall glucose challenge test threshold for the Carpenter and Coustan criteria is not 7.8 mmol!L but 7.3 mmol!L. As noted above, the latter threshold leads to a sensitivity of 80.4% and a specificity of 74.6%, but the false-positive rate is 23.6%, almost double that which occurs with the National Diabetes Data Croup criteria and glucose challenge test thresholds adjusted for time since last meal. Mter the analysis to calibrate new time-specific thresholds for the glucose challenge test against the Carpenter and Coustan oral CIT is repeated, the false-positive rate is even higher, at 25.5%, although the false-negative rate has dropped to 0.9% from 1.2%. Thus the advantages of adjustment for time since last meal are much smaller with the Carpenter and Coustan criteria. In either case there is a definite increase in numbers of oral CITs that will be performed if practitioners use the Carpenter and Coustan criteria and wish to maintain a high level of screening sensitivity with the glucose challenge test. Some practitioners may believe that detection of gestational diabetes is so important that good or excellent sensitivity should alone determine the choice of glucose challenge test threshold values. To that end, we have also reported threshold values for sensitivities closest to 80% and 90% for both sets of criteria, with

Sermer et al.

Volume 171, Number 3 Am J Obstet Gynecol

615

Carpenter and Coustan criteria Last meal <1 hr

1-2 hr

2-3 hr

>3 hr

Overall

Cut point "<'!7.3 mmol/L

Cut point "<'!6.6 mmol/L

Cut point "<'! 7.1 mmol/L

Cut point "<'!6.8 mmol/L

Cut point "<'!7.2 mmol/L

Cut point "<'!7.0 mmol/L

Cut point "<'!8.0 mmol/L

Cut point "<'!7.0 mmol/L

Cut point "<'!7.3 mmol/L

Cut point "<'!6.9 mmol/L

82.3 75.0 23.3 1.2

91.1 58.7 38.5 0.6

83.5 75.6 22.7 1.1

89.9 69.0 28.9 0.7

79.2 72.2 25.9 1.4

86.8 67.5 30.3 0.9

77.8 77.9 20.6 1.5

90.7 57.6 39.5 0.6

80.4 74.6 23.6 1.4

89.4 65.8 31.8 0.7

and without subgroups by time smce last meal. Attempts to increase sensitivity and minimize false-negative results lead to lowered specificity, with clear increases in the number of oral GITs performed, a phenomenon that is particularly pronounced with the Carpenter and Coustan criteria. This study demonstrates that the time since last meal has a significant impact on the observed plasma glucose levels when gravid patients undergo a routine glucose challenge test. In turn, this alters the optimum threshold for determining which patients should go on to the definitive oral GIT. We suggest that glucose challenge test threshold values of 8.2 mmoVL should be used if the patient ate 2 hours before the screening test, whereas threshold values of 7.9 mmoVL and 8.3 mmoVL should be used if patients ate, respectively, between 2 and 3 hours and > 3 hours before the glucose challenge test. This very simple change in policy would lead to improved efficiency of the screening test, decreased frequency with which patients require the inconvenient 3-hour oral GIT, and decreased direct and indirect health care costs totaling millions of dollars per year. The adoption of these cutoff values will, however, need to be reconsidered if the oral GIT thresholds, as defined' by the National Diabetes Data Group, are no longer used as the standard for diagnosing gestational diabetes mellitus. We thank our study coordinators Michelle Bailey, Vicki Bowden, and Elizabeth Peloso, Ms. Caroll Gomez of the Sunnybrook Clinical Epidemiology Program, and Ms. Ruth Milner of the Vancouver Children's Hospital for advice and support in data management and statistical analysis. The following are the Toronto Tri-Hospital Gestational Diabetes Investigators: D. Pond, MD, H. Khosid, MD, I. Eisen, MD, S. Shafir, MD, R. Livingstone, MD, S. Plosker, MD, L. Tanzer, MD, B. Cruickshank, MD, A. Karlinsky, MD, E. Hussman, MD, W. Wolfman, MD, D. Giannoulias, MD, T. Doran, MD, B. Thomas, MD, E. Herer, MD, E. Woloschuk, MD, E. Lyons, MD, J. Greenberg, MD, P. Bernstein, MD,

J. Johnson, MD, A. Bernstein, MD, R. Perlin, MD, B. Jackson, MD, A. ZaItz, MD, J. Rudderman, MD, F. Engle, MD, T. Reid, MD, K. Chu, MD, I. Goldman, MD, M. Barkin, MD, H. Ho, MD, K. Amankwah, MD, R. Springate, MD, P. Hose, MD, B. Moran, MD, D. Robb, MD, N. Braithwaite, MD, D. Ruddock, MD, N. Hughes, MD, S. Gottesman, MD, P. Hawrylyshyn, MD, B. Newman, MD, R. Brooks, MD, B. Ludwig, MD, F. Weisberg, MD, L. Philip-White, MD, F. Boucher, MD, P. Freedman, MD, H. Grundman, MD, L. Wilson, MD, K. Valtchev, MD, Y. Yung, MD, F. Harris, MD, A. Egier, MD, F. Papsin, MD, S. Nusinowitz, MD, B. Marlow, MD, J. Bean, MD, J. Marmoreo, MD, S. Hanet, MD, and M. Pearl, MD. REFERENCES 1. Summary and recommendations of the Second International Workshop Conference on gestational diabetes mellitus. Diabetes 1985;34(suppl 2):123-6. 2. Freinkel N, Metzger BE, Phelps RL, et al. Gestational diabetes mellitus: heterogeneity of maternal age, weight, insulin secretion, HLA antigens, and islet cell antibodies and the impact of maternal metabolism on pancreatic B-cell and somatic development in the offspring. Diabetes 1985;34(suppI2):1-7. 3. Gabbe SG, Mestman JH, Freeman RK, Anderson GY, Lowensohn RI. Management and outcome of class A diabetes mellitus. AM J OBSTET GYNECOL 1977;127:465-9. 4. Coustan DR, Imarah J. Prophylactic insulin treatment of gestational diabetes reduces the incidence of macrosomia, operative delivery, and birth trauma. AM J OBSTET GYNECOL 1984;150:836-42. 5. Kitzmiller JL, Cloherty JP, Younger MD, et al. Diabetic pregnancy and perinatal morbidity. AM J OBSTET GYNECOL 1978; 131 :560-80. 6. Widness JA, Cowett RM, Coustan DR, Carpenter MW, Oh W. Neonatal morbidities in infants of mothers with glucose intolerance in pregnancy. Diabetes 1985;34(suppI2):61-5. 7. O'Sullivan JB, Gellis SS, Dandrow R, Tenney R. The potential diabetic and her treatment in pregnancy. Obstet Gynecol 1966;27:683-9. 8. O'SullivanJB, Mahan CM, Charles D, Dandrow RV. Medical treatment of the gestational diabetic. Obstet Gynecol 1974;43:817-21. 9. Pettitti DJ, Bennett PH, Knowler WC, Baird HR, Aleck KA. Gestational diabetes mellitus and impaired glucose tolerance during pregnancy: long-term effects on obesity and glucose tolerance in the offspring. Diabetes 1985;34(suppl 2):119-22.

616

Sermer et al.

September J 994

Am

10. American Diabetes Association. Position statement on gestational diabetes mellitus. A,,,, J OBSTET GYNECOL 1986; 156: 488-9. 11. Lind T, Hytten FM. The excretion of glucose during normal pregnancy. J Obstet Gynaecol Br Commonw 1972; 79:961-5. 12. Davies DM, Welborn 'fA. Glycosylated haemoglobin in pregnancy. Aust N Z .I Obstet Gynecol 1980;20: 147 -50. 13. Cousins L, Dattel BJ, Hollingsworth DR, Zettner A. Glycosylated hemoglobin as a screening test for carbohydrate intolerance in pregnancy. AM J OBSTET GYNECOL 1984; 150: 455-60. 14. Shah BD, Cohen AW, May C, Gabbe SG. Comparison of glycohemoglobin determination and the one-hour oral glucose screen in the identification of gestational diabetes. AM .I OBsTn GVNECOI. 1982;144:774-6. 15. Artal R, Mosley GM, Dorey FJ. Glycohemoglobin as a screening test for gestational diabetes. A,,,, J OBSTET GYNECOL 1984;148:142-3. 16. Mc Farland KF, Murtiashaw M, Baynes JW. Clinical value of glycosylated serum protein and glycosylated hemoglobin levels in the diagnosis of gestational diabetes mellitus. Obstet Gynecol 1984;64:516-8. 17. Lind T. Antenatal screening for diabetes mellitus. Br .I Obstet Gynaecol 1984;91:833-4. 18. Lind'C McDougall AN. Antenatal screening for diabetes mellitus by random blood glucose sampling. Br J Obstet Gynaecol 1981 ;88: 346-51. 19. Lind T. Antenatal screening using random blood glocose values. Diabetes 1985;34(suppl 2):17-20. 20. O'Sullivan JB, Mahan CM, Charles D, Dandrow RV. Screening criteria for high-risk gestational diabetic patients. AM .I OllsTn GVNECOL 1973; 116:895-900. 21. O'Sullivan JB. Establishing criteria for gestational diabetes. Diabetes Care 1980;3:437-9. 22. Carpenter MW, Coustan DR. Criteria for screening tests for gestational diabetes. A,,,, .I OBSTET GVNECOL 1982; 144: 768-73.

J Obstct Gynecol

23. Coustan DR, Carpenter MW. Detection and treatment of gestational diabetes. Clin Obstet Gynecol 1985;28:507-15. 24. Jovanovic L, Peterson CM. Screening for gestational diabetes: optimum timing and criteria for re-testing. Diabetes 1985;34(suppl 2):21-3. 25. Coustan DR, Widness .lA, Carpenter MW, Rotondo L, Pratt DC, Oh W. Should the fifty-gram, one-hour plasma glucose screening test for gestational diabetes be administered in the fasting or fed state? AM J OBSTET GVNECOL 1986; 154: 1031-5. 26. Dietrich ML, Dolnicek TF, Rayburn WF. Gestational diabetes screening in a private, midwestern American population. AM J OBSTET GYNECOL 1987;156:1403-8. 27. Dooley SL, Ross IC. In discussion. Dietrich ML, Dolnicek TF, Rayburn WF. Gestational diabetes screening in a private, midwestern American population. AM .I OBSTET GVNECOL 1987; 156: 1406-1407. 28. Carpenter MW, Coustan DR. Criteria for screening tests for gestational diabetes. AM .I OSSH:T GYNECOI. 1982;144: 768-73. 29. Magee MS, Walden CE, Benedetti TJ, Knopp RH. Influence of diagnostic criteria on the incidence of gestational diabetes and perinatal morbidity . .lAMA 1993;269:609-15. 30. National Diabetes Data Group. Classification and diagnosis of diabetes mellitus and other categories of glucose intolerance. Diabetes 1979;28: 1039-57. 31. Sacks D, Abu-Fadil S, Greenspoon .IS, Fotheringham N. Do the current standards for glucose tolerance testing in pregnancy represent a valid conversion of O'Sullivan's original criteria? AM J OBSTET GYNECOL 1989; 161 :638-41. 32. Hanley JL, McNeil BJ. The meaning and use of the area under the receiver operating characteristic (ROC) curve. Radiology 1982;143:29-36. 33. Hanley .lA, McNeil BJ. A method of comparing the areas under receiver operating characteristic curves derived from the same cases. Radiology 1983; 148:839-43.

Availability of JOURNAL Back Issues As a service to our subscribers, copies of back issues of the AMERICAN JOURNAL OF OBSTETRICS AND GYNECOLOGY for the preceding 5 years are maintained and are available for purchase from the Publisher, Mosby-Year Book, Inc., at a cost of $10.50 per issue. The following quantity discounts are available: one fourth off on quantities of 12 to 23 and one third off on quantities of 24 or more. Please write to Mosby-Year Book, Inc., Subscription Services, 11830 Westline Industrial Drive, St. Louis, MO 63146-3318 or call (800) 453-4351 or (314) 453-4351 for information on availability of particular issues. If back issues are unavailable from the Publisher, photocopies of complete issues are available from University Microfilms International, 300 N. Zeeb Road, Ann Arbor, MI 48106. Tel.: (313) 761-4700.