- Email: [email protected]
Accurate prediction of duodenal-ulcer healing rate by discriminant analysis
GASTROENTEROLOGY
1983:85:403-12
Accurate Prediction of Duodenal-Ulcer Healing Rate by Discriminant Analysis SHIU-KUM
LAM
and JARLEY
KOO
Combined Gastrointestinal Unit, Departments Kong, Queen Mary Hospital, Hong Kong
Previous studies have shown that -50% and 70% of duodenal ulcers heal after 2 and 4 wk, respectively, of cimetidine, and at least one-third heal after 4 wk of placebo. In order to identify these groups of ulcers before treatment, a two-phase study was pe$ormed, including an initial double-blind trial of cimetidine vs. placebo in 120 patients, and a subsequent open study with identical protocol of cimetidine vs. no cimetidine in another 60 patients. Forty clinical, personal, physiologic, and endoscopic characteristics were prospectively obtained in each patient, and were analyzed by stepwise discriminant analysis at the end of phase 1. This identified the discriminants against healing after 2 wk of cimetidine as late onset disease, body weight, and ulcer diameter; those after 4 wk of cimetidine as analgesic consumption, neurosis, low fasting serum gastrin, low pentagastrin D 5o and ulcer diameter; and those after 4 wk of placebo as back pain, bleeding, and alcohol consumption. Based on the discriminant scores derived, the sensitivity, specificity, and eficiency of prediction for complete healing as determined endoscopically were 74.4% 90% and 82.3% for 2-wk cimetidine, lOOoh, 87.5% and 97.5% for 4wk cimetidine, and 85.7% 83.3% and 84.2% for 4wk placebo treatment. In phase 2, correct predictions were made in 36 of 40, 39 of 40, and 17 of 20 patients treated, respectively, for 2 and 4 wk with cimetidine, and 4 wk without cimetidine. Accurate Received July 19, 1982. Accepted March 3, 1983. Address requests for reprints to: Dr. S.-K. Lam, M.D., Department of Medicine, University of Hong Kong, Queen Mary Hospital, Hong Kong. This work was supported in part by University Research Grants (311/030/8009/31, 335/041/9908) and Wing Lung Bank Medical Research Fund (335/041/9967), University of Hong Kong. The authors acknowledge the expert statistical advice of Janet Elashoff. They are grateful to S. F. Pang, W. H. Tang, and J. Yu of the Computer Center, University of Hong Kong, for their computer assistance; M. Chong, D. Cheung, and B. Ng for nursing assistance; P. Yip and G. H. Joe for technical help; J. Watt for secretarial help; and Smith Kline & French Laboratories for sponsorship. 0 1983 by the American Gastroenterological Association 0016-5085/83/$3.00
of Medicine
and Surgery,
University
of Hong
prediction of duodenal-ulcer healing rate with and without cimetidine is thus possible by discriminant analysis. As many medical and surgical modalities of treatment are now available, this approach should have the potential of selecting the appropriate form of treatment for a given patient. The present double-blind controlled trial comparing placebo with two doses of cimetidine, namely 0.5 g and 1 g daily, in the healing of duodenal ulcer was performed to address two unresolved questions. First, it has been found that in a geographically variable group, -33.3%-66.6% of placebo-treated patients and an average of 70% of cimetidine-treated patients with duodenal ulcers would heal after 4-6 wk of treatment with cimetidine (1,2), and >50% would heal after only 2 wk of treatment (3). If it becomes possible to accurately predict which patients will heal or not heal after 4 wk of cimetidine treatment, which will heal after only 2 wk of cimetidine treatment, and which will heal with 4 wk of placebo treatment, this will not only be of therapeutic and economic importance, but will help to elucidate the factors that affect the healing of duodenal ulcer as well as the factors that underlie its pathophysiology. Because duodenal ulcer is most probably heterogeneous in etiology as evident clinically (4-6), genetically (7-9), and pathophysiologically (lO,ll), no single factor is likely to discriminate ulcers that heal from those that do not heal with treatment. A composite approach encompassing as many factors as are known may be the answer. This study prospectively, and as comprehensively as possible, collects the factors that might pertain to healing, including clinical, personal, physiologic, and endoscopic characteristics. In phase 1, which comprised the trial itself, all of these characteristics were subjected to stepwise discriminant analysis for healing and nonhealing. In phase 2, the accuracy of prediction was tested in a new group of patients who went through the same protocol, except that they
GASTROENTEROLOGY
404 LAM AND KOO
were randomly treated with or without cimetidine. Second, compared with whites, Chinese patients with duodenal ulcer had significantly smaller parieta1 cell mass and histamine- or pentagastrin-stimulated maximal acid output (12-14). Assuming that the adequacy of acid reduction by cimetidine depends on (a] the size of the parietal cell mass and (b) the dose of cimetidine used, it is possible that smaller doses of cimetidine than those established in whites may be adequate for the healing of duodenal ulcer in Chinese. Indeed, a recent clinical trial indicated that antacid tablets with a neutralizing capacity of only 12.5mmol HCl/tablet, given in tablet form 1 and 3 h after the three main meals of the day and at bedtime, healed 76.9% duodenal ulcers compared with 33.3% using placebo (15). If a substantially smaller dose of cimetidine proves effective, it would mean that cimetidine would become economically accessible to substantially more patients in the developing Asian countries, and would also suggest that the dose of cimetidine may be reduced in patients with small acid secretion.
Methods Patients Candidates were recruited from patients attending the University Medical and Surgical Gastrointestinal Clinics, as well as patients admitted to the general medical, surgical, and gastrointestinal wards of the University Departments of Medicine and Surgery, Queen Mary Hospital, Hong Kong. This has the advantage of including a wide spectrum of patients in the study. Candidates were excluded from consideration if they had a concomitant medical problem, particularly renal diseases, cardiovascular diseases, diabetes mellitus, or chronic obstructive airways diseases, pyloric stenosis, previous gastric operations for ulcer, or an associated gastric ulcer. The protocol of the study was approved by the Committee on Higher Degrees and Research Grants, University of Hong Kong.
Clinical and Personal
Characteristics
After being accepted into the study, the patients were interviewed in detail and the clinical and personal characteristics as listed later in Table 2 were carefully recorded. Further explanation is necessary regarding the following points: early-onset patients were defined as those having the onset of ulcer symptoms at 530 yr of age, and late-onset as those having the onset at >30 yr of age. The cutoff point at age 30 yr was based on previous studies indicating that early-onset patients differed from late-onset patients, particularly in being more familial, having the same blood group distribution as the general population instead of blood group 0 predominence, as in late-onset patients (5,16), and having higher meal-stimulated gastrin response (17). Symptom period was the number of days before the study during which the patient suffered from the current pain. Pain severity, which was also assessed at
Vol. 85. No. 2
subsequent visits, was scored as follows: nil = 0; mild = 1; off-work or work-interfering = 2; and requiring immediate medical attention (e.g., emergency room services) = 3. Pain duration was the average duration of the current pain in hours. Familial dyspepsia was considered present when the patient indicated that ulcer dyspepsia, gastrointestinal bleeding, perforation, or gastric surgery, or a combination of any or all, had occurred in a first-degree relative. This was scored as nil = 0, one first-degree relative = 1, two first-degree relatives = 2, and three or more first-degree relatives or both parents = 3. Stress at work was scored subjectively by the patient as 0-3, and included work pressure or labor intensity of the job, or both. Neurosis was scored according to the number of neurotic traits present with a maximum of 3 (insomnia, tension headache, palpitation, hand tremors, nonpostural dizzy spells). Habitual cigarette smoking was scored as nil = 0, %lO/day = 1, and >lO/day = 2. The majority of smokers had smoked continuously for >lO yr, and all smokers had this habit for at least 5 yr. Habitual alcohol consumption for at least 5 yr was scored as nil = 0, estimated alcohol ~50 g/day = 1, >50 g/day = 2. Habitual analgesic consumption for at least 1 yr was scored as nil = 0, 1 dose/m0 = 1, at least 1 doselwk = 2; analgesics included aspirin, nonsteroidal antiinflammatory agents, and the many forms of patent Chinese herbal medicine, the nature of which was mostly undefined. The patients were not told to stop their smoking, alcoholic, or analgesic habits, but were told that these might be harmful to ulcer healing. At subsequent visits, these habits were assessed by direct questioning and the consumption per day was recorded. The consumption during treatment was scored at the end of the trial as nil to begin with = 0, withdrawn completely = 1, reduced by >50% = 2, and unchanged = 3. In addition, before and at each subsequent visit, the following nonulcer symptoms were recorded: diarrhea, constipation, skin rash, dizziness, headache, palpitation, blurring of vision, dryness of mouth, gynecomastia or galactorrhea, impotence, and libido (frequency of sexual intercourse over the past week). Each patient was also asked to record the occurrence of pain, its severity, and the number of antacid tablets consumed, on a tabulated diary card.
Medications Two doses of cimetidine, 1 g/day and 0.5 g/day, were chosen to examine the hypothesis that Chinese patients required a substandard dose of cimetidine. Since some of the potential discriminating factors for healing may be directly or indirectly related to acid secretion and, therefore, acid inhibition, the use of full and half doses of cimetidine may have an added advantage of bringing out the effects of these factors on healing. Cimetidine tablets, 100 mg or 200 mg, or placebo (Smith Kline & French Laboratories, Philadelphia, Pa.), all identical in appearance and taste, were packaged in bottles containing weekly supplies of 35 tablets and carrying a code number and printed instructions. There were four bottles under each code number. Each patient was given one bottle at each visit and was asked to take one tablet immediately before breakfast, lunch, and supper, and two tablets before retir-
August 1983
PREDICTIONOF DUODENAL-ULCER HEALING 405
ing at night. At each visit, each patient was also given 40 antacid tablets [Gelusil (Parke-Davis, Santurce, P.R.] containing magnesium trisilicate 500 mg and dried aluminium hydroxide gel 250 mg with a neutralizing capacity of 6 mEq of HCl], and was asked to chew one or two tablets for pain relief. Patients were instructed to eat no more than three regular meals per day, and not to take any other medicine without notifying the trial clinic sister. They were told to bring along at each visit any remaining medications, which would then be counted. Physiologic
Low DSoc previously in 100 normal Chinese subjects (14). was arbitrarily defined as
Characteristics Endoscopic
On two separate mornings before the commencement of treatment, the following two tests were performed in random order after fasting from midnight. After a blood sample was taken for fasting serum gastrin, the patient ate a standard meal of 50 g of protein, 40 g of carbohydrate, and 40 g of fat in the form of a milk drink [60 g of Sustagen (Mead Johnson & Company, Evansville, Ind.) in 60 ml of milk] and Chinese-style rice congee (130 g of minced beef in 240 ml of water and rice with 5 ml of cooking oil). Further blood samples were taken at 30, 60, and 120 min after the start of the meal. Serum from all samples was immediately extracted by centrifugation and stored at - 20°C. Gastrin estimations by radioimmunoassay were done in duplicate as previously detailed (18). Sensitivity was below 2.5pmol/L of serum. Within and between assay coefficients of variation were 6.2% and 13.7%,respectively. The antiserum reacted equally with G-17 and G-34 gastrin. The integrated gastrin response after the standard meal of each subject was calculated by computing the total area under the response curve from 0 to 120 min. On a separate day, gastric acid output in response to infusions of pentagastrin administered as follows was measured: 0 (0.15M saline, 1.5mmol/h), 93.8, 375,1500, and 6000 nglkg * h. Each dose of pentagastrin was infused for 30 min into a forearm vein by a syringe infusion pump (Harvard Apparatus Co., Inc., Millis, Mass.). The final dose, 6000 ng/kg * h, was infused for 60 min. Gastric juice was collected at 15-min intervals. The acid output during saline infusion was the basal acid output (BAO) and that was the during infusion of 6000 ng/kg - h pentagastrin maximal acid output (MAO). The dose required for halfmaximal response (DsO)was estimated by linear regression analysis according to the equation: response = calculated maximal response - DsO (response/dose), which gave Dso as the negative slope. The D5,, corrected for basal acid output (DsoC) was calculated according to the equation: D5,,c = D5,, [l - basal acid output/(calculated maximal acid output + basal acid output)]. The DsOc was taken to represent the sensitivity of the parietal cells in response to pentagastrin stimulation; the methodologies, validation experiments, and calculations were detailed previously (19,20). Based on the above, the physiologic characteristics, as listed later in Table 2, were bbiainedy A hypersecretor was defined as a patient having MAO/kg total body wt (correction for body weight reduces differences between the two sexes) above 2 SD of the mean of normal as established
Characteristics
Endoscopy was performed with a forward-viewing panendoscope (Olympus GIF-P, Olympus Corp. of America, New Hyde Park, N.Y.) within 72 h before the commencement of treatment, and at the end of 2 and 4 wk of treatment, with each patient endoscoped throughout the study by the same author who was unaware of the patient’s form of treatment and clinical progress. The diameters and depth of the ulcer were estimated using the tip of the biopsy forceps. Patients were excluded from the study if the longest diameters of their ulcers were <3 mm or >25 mm. The endoscopic characteristics, as listed later in Table 2, were recorded. Ulcer site was described as anterior, posterior, floor, roof, and apex (near the exit) of the duodenal bulb. The degree of surrounding inflammation was scored as gross hyperemia, granularity, or edema (or any combination thereof) = 3, moderate = 2, mild = 1, absent = 0. The degree of deformity of the duodenal bulb was scored: actual narrowing of entrance or exit of the duodenal bulb by scar but still admitting the endoscope (external diameter = 9 mm] = 3, bilateral deformity = 2, unilateral deformity of bulb = 1,absent = 0.The findings at the end of 2 and 4 wk of treatment were compared with those observed at the initial endoscopy, and the patient was classified immediately as healed (complete disappearance of ulcer with or without residual inflammation usually in the form of hyperemia and granularity) or unhealed (persistance of ulcer, however small). Statistical
Analysis
On the assumption that 70% of cimetidine-treated patients and 35% of placebo-treated patients would heal after 4 wk, we calculated that we should include 37 subjects in each group to show a difference of 35% with a type I error of 0.05 and a type II error of 0.15(one-sided test) (21). Trial medications were prepared so that each treatment group consisted of 40 patients. Chi-square test with Yates’ correction, one-way analysis of variance, and paired t-test were used as appropriate (22). Values were expressed as mean + SE, and p values <0.05 were considered significant. Study
Design
Phase 1. This was the double-blind controlled study itself, as previously detailed, and consisted of 4 wk of treatment with endoscopic assessment at the end of the 2nd and 4th week.
406 LAM
(;r\S'rKOEN1‘EKOI,O(;\r' L'ol. 85 No.
AND K00
Phase 2. A further series of 60 patients were recruited by the same criteria. They were randomly allocated to two forms of treatment in a proportion of 2 : 1. These were, respectively: (a) cimetidine 1 g/day (200 mg t.d.s. immediately before meals and 400 mg at bedtime) and (b) Gelusil tablets (one or two to be chewed for pain relief). The protocol was otherwise exactly the same as in phase 1.
Stepwise
Discriminant
2
sal Ine
Analysis
At the end of the phase 1 study, patients were divided into those who had received cimetidine (0.5 glday or 1 g/day) and those given placebo. Three groups of patients, namely those who had either 2-wk cimetidine, 4wk cimetidine, or 4-wk placebo, were subjected to stepwise discriminant analysis with healed ulcer as the dependent variable and the characteristics listed later in Table 2 as the independent variables, using the Statistical Package for the Social Sciences (SPSS) programs (23). The computer employed was the IBM OS/VSl.IPO (IBM, White Plains, N.Y.). The criterion used was Rao’s V. This method maximized separation of healed and unhealed groups by weighing and combining the discriminating variables in a linear form. The single best discriminant variable was first selected, the remaining potential variables were then tested, and the variable that best improved the criterion (Rao’s V) in combination with the first variable was then chosen. Subsequent variables were selected according to their contributions to further discrimination, and previously selected variables were reexamined at each step. If a variable reduced discrimination when combined with subsequently selected variables, it was removed, but may have been subsequently reentered if proven again to be of value. Variables were thus added until additional variables no longer contributed significantly (p < 0.05) to further discrimination (change in Rao’s V). The standardized and unstandardized canonical discriminant function coefficients were calculated for the selected variables. From the standardized coefficients, the relative contribution of the variables in discriminating patients with healed ulcer from those with unhealed ulcer could be realized (vide infra). From the unstandardized coefficients, the discriminant score for each patient could be calculated (vide infra), and the patients were classified based on probability into “healed” and “unhealed” membership groups. To avoid overoptimistic bias in the final discriminant function, a conditional rule was used in the stepwise selection of variables, so that their maximum F-TO-ENTER and their minimum F-TO-REMOVE carried a p value of at least 0.05 (i.e., p IN and p OUT ~0.05).This would yield the least number of variables with the best discriminant function (24). In the discriminant analysis for the 4-wk placebo treatment, however, no variable was selected by this stringent rule, and the p IN and p OUT values were relaxed to 0.1 for that particular analysis. The accuracy of the prediction of healing was tested by asking the computer to classify the phase 2 patients into healed and unhealed, based on the selected discriminants derived from the phase 1 study. Information concerning healing of the phase 2 patients was withheld from the computer.
1
0
93.9
1
I
375
1500
6000
6000
PENTAGASTRIN INFUSION ng.Kg-'.h-' Figure
I. Pentagastrin dose-response studies after bolus injections of saline, 100 mg of cimetidine, or 200 mg of cimetidine in 6 duodenal ulcer patients. No significant differences between 100 mg and 200 mg cimetidine. Mean t SEM.
Results Cimetidine
and
Pentagastrin
Dose Response
Significant inhibition in acid output was observed after administration of 100 mg or 200 mg cimetidine. No differences in acid output were observed after administration of 100 mg and 200 mg cimetidine (Figure 1). Mean DsUC (nanograms per hour per kilogram) on the saline day [24.4 t 5.6 (SE)] was significantly lower (p < 0.03) than that after 100 mg cimetidine [289.5 2 91.9 (SE)] and after 200 mg cimetidine (391.2 ? 118.1). There was no significant difference
between
100 mg
and
200
mg cimetidine
(p < 0.3).
Phase
1 Study
group Patients. Two patients in the placebo and one in the 0.5 glday cimetidine group defaulted. One in each became asymptomatic and refused subsequent endoscopies, and the other left to attend to urgent business in China. No significant difference in each of the 40 patient characteristics listed in Table 2 could be detected among patients in the placebo group, 0.5 g/day cimetidine group, and 1 g/day cimetidine group (F ratio ranged between 0.01 and 2.3,x2 between 0.4 and 1.1, p values between 0.9 and 0.1). Some of these characteristics are presented in Table 1. The mean number of “cimetidine” tablets omitted per week in the placebo group, 0.5 g/day cimetidine group, and 1 g/day cimetidine group were 2.0 + 0.5, 2.0 -+ 0.6, and were mean
2.3 ? 0.7, respectively,
not significantly different number of Gelusil tablets
and
from each other. The taken per day in the
first 2 wk was not significantly different among these treatment groups (1.9 * 0.4, 2.0 ? 0.5, and 2.8 -t 0.9,
August 1983
Table
PREDICTION OF DUODENAL-ULCER HEALING
1. Comparison of Pretrial Daily Cimetidine, and
Treated Characteristics As WeI1 As Healing Rates of Patients 1 g Daily Cimetidine in the Phase 1 and Phase 2 Studies
With Placebo,
Phase 1 Placebo Total No. of patients Age (yr, range) Male Duration of symptom (yr) Familial dyspepsia, positive n scorea Blood group 0 No. with previous bleeding Cigarette smokers Alcohol users Analgesic users BAO (mmolih) MAO (mmol/h) DX (ng/kg * h) Fasting gastrin CfmoI/ml) Postprandial gastrin (nmol.min/L) Ulcerdiameter(mm) Ulcer depth (mm) Ulcer healing % healed z wk vs. placebo (p) % healed 4 wk vs. placebo (p)
Cimetidine (0.5g/day)
407
0.5 g
Phase z Cimetidine (1 g/day)
P
Cimetidine (1 g/day1
No cimetidine
38 44.0(16-75) 28 9.22 1.9 22 1.72 0.1 16 22 21 6 5 3.6f 0.5 25.5-+1.3 139.1k 26.0 36.7t 5.8 9.4k 1.0
47.2(19-75) 29 8.5* 1.3 19 1.9f 0.2 17 26 20 8 4 2.72 0.3 25.3-t1.7 129.9* 19.8 39.0? 4.9 10.6? 1.8
40 48.6(19-73) 28 11.5+ 1.9 23 1.82 0.2 24 23 19 5 6 2.5k 0.3 25.3k 1.6 168.4+ 29.1 32.02 8.3 7.8+ 0.5
CO.5 Cl.0
40 41.1(20-68) 27 9.1" 1.6 22 1.8f 0.2 21 23 19 6 6 3.0+ 0.4 24.8k 1.7 143.1+ 22.5 40.3+ 3.8 9.8" 1.2
20 43.5(25-64) 15 11.12 2.4 9 1.8f 0.1 11 12 10 3 3 2.8f 0.2 26.2f 1.3 140.42 23.0 36.42 3.9 8.2k 1.1
7.42 0.6 2.1+ 0.1
7.1+ 0.6 2.2t 0.1
7.2t 0.6 2 t 0.1
CO.9 CO.5
6.92 0.7 1.9k 0.1
7.25 0.6 2.0* 0.1
10.5
48.7 <0.0005 82.1 <0.0005
52.5
10.0
<0.0005 77.5 <0.0005
82.5
30.0
36.8
39
50.0
BAO = basal acid output. MAO = maximal acid output. D5~ = corrected dose required for half-maximal response. No differences in healing rates between the two cimetidine groups were observed at 2 wk (p < 0.9) and 4 wk (p < 0.5). No differences among the three cimetidine groups and between the placebo and no cimetidine group were observed. Mean 2 SEM. ’ See Methods section.
respectively), but were significantly (p < 0.02) more in the placebo (2.2 + 0.5) than in the 0.5 g/day cimetidine (0.5 + 0.2) or in the 1 g/day cimetidine (0.5 + 0.3) group in the second 2 wk. No gynecomastia, galactorrhea, impotence, change in libido, or hematologic or biochemical abnormality was recorded in any patient. Minor nonulcer symptoms (vide supra) were noted as new events in 15.8%, 15.4%, and lo%, as old complaints in 26.3%, 30.7%, and 30%, and to be better in 10.5%, 10.3%, and 12.5% of patients treated, respectively, with placebo, 0.5 g/day cimetidine, and 1 g/day cimetidine. There were no significant differences among groups. Healing. Both 0.5 g/day and 1 g/day cimetidine healed significantly more ulcers than placebo after 2 or 4 wk of treatment, and their healing rates were not significantly different from each other. In relation to these findings, it is pertinent to note that the corresponding healing rates of patients treated with and without cimetidine in the phase 2 study were similar to those observed in phase 1 (Table 1). After 2 and 4 wk of treatment, 29.3% and 7.3% of patients with healed ulcer, respectively, remained symptomatic, whereas 75% and 74.3% of patients with unhealed ulcer, respectively, were asymptomatic.
Individual analysis. Compared with those who healed, patients who did not heal after 2 wk of treatment were significantly older at the onset of symptoms, and had a shorter symptom period and greater body weight. More of them were cigarette smokers who smoked more heavily and who abstained less during treatment. More of them were analgesic users. More had greater ulcer diameter and depth, less posterior and more apical ulceration, more multiple ulceration, and more inflammation. The group that did not heal at the end of 4 wk contained significantly more cigarette smokers who smoked more heavily and significantly more subjects with greater ulcer diameter and more apical ulceration. A greater proportion in this group had low DSO in the pentagastrin dose-response test. Because multiple comparisons were made (n = 40), the p values were corrected by multiplying by a factor of 40. When this was done, only cigarette smoking (p < 0.002), number of cigarettes smoked (p < 0.011, ulcer diameter (p < O.OOZ), and ulcer depth (p < 0.02) remained significant (Table 2). A significant correlation was observed between cigarette consumption and ulcer diameter (Figure 2). Discriminant analysis. The selected discriminant for healing as based on patients in the
408
LAM AND
Table
2.
CASTROENTEROLOCY
KOO
Contribution
of Various
Characteristics
to Healing
as Analyzed
Individually
by x2 or Analysis
Individual values
Clinical Onset age Early/late onset, early onset Duration of symptoms Remission period Symptom period Pain duration Pain score Nocturnal awakening pain Back pain Familial dyspepsia, positive strength Bleeding, positive Personal Sex
phase
are expressed
29.7% 60.8% 0.9 k 0.1 59.5"/,
0.70 0.30 0.60 0.90
24.7”A 50.6% 0.9 2 0.1 62.30,
32.5% 62.5% 1.1 + 0.2 57.5%
0.15 0.10 0.30 0.80
72% 42.8 2 2.6 50.7 2 1.3 51.2"/, 30.2"/, 0.6 + 0.1 0.8 + 0.2 14.0% 0.1 + 0.1 0.2 + 0.1
o-3 o-3 o-3
4.701 0.1 + 0 0.3 k 0 1.4 + 0.2 0.7 + 0.1
73% 48.8 I? 1.9 54.1 5 0.9 47.3"/, 63.50, 1.1 2 0.3 1.4 + 0.2 17.60, 0.3 + 0.1 0.2 2 0.1 17.6"h 0.2 k 0.1 0.3 ? 0.1 1.1 k 0.1 0.7 lr 0.1
0.70 0.06 0.04 0.80 0.0005 0.003 0.02 0.80 0.04 0.30 0.03 0.03 0.30 0.20 0.90
70.1% 46.8 ‘- 1.8 52.7 + 0.9 54.5% 42.8”/, 0.8 + 0.1 1.1 t 0.2 13.0% 0.2 + 0.1 0.2 t 0 9.1% 0.1 + 0 0.3 2 0.1 1.2 + 0.1 0.6 + 0.1
77.5% 46.4 + 2.9 53.6 + 1.4 37.5% 67.501 1.2 k 0.1 1.4 + 0.2 22.5”Aj 0.4 -t 0.1 0.2 + 0.1 200, 0.3 -+ 0.1 0.3 2 0 1.2 2 0.2 0.9 -’ 0.2
mmolih mmol/h 1 = no, 2 = yes ng/kg . h 1 = no, 2 = yes pmol/L nmol * min/L
2.8 + 0.4 25.4 " 1.6 48.8% 158.1 2 22.5 46.S"h 31.6 k 2.7 9.7 t 0.9
3.0 k 0.3 25.2 5 1.1 62.2% 138.5 + 18.7 60.8% 38.3 + 3.7 9.0 + 1.0
0.50 0.70 0.20 0.80 0.10 0.30 0.55
2.8 + 0.3 25.7 + 1.2 55.8”/, 162.2 -+ 17.8 46.8% 37.8 + 3.1 9.4 + 1.0
3.2 k 0.4 24.9 k 1.5 6001 116.0 + 26.0 72.5% 32.2 k 5.3 9.0 Zk 1.0
0.40 0.65 0.50 0.20 0.04 0.60 0.70
48.8% 20.9% 16.2% 4.9"y" 2.3% 5.5 k 0.3 1.8 k 0.1 7.0% 1.2 2 0.1 0.7 2 0.1
41.9% 6.801 27% 10.8”/, 16.2”r a.2 ? 0.5 2.3 k 0.1 16.2% 1.5 + 0.1 0.6 + 0.1
49.3% 14.301 18.2% 10.4% 5.2% 6.6 k 0.4 2.0 2 0.1 12.3”/r, 1.5 +- 0.1 0.6 -c 0.1
35% 7.5% 32.501 5% 22.5% 8.5 ” 0.7 2.3 2 0.1 13.9”/, 1.3 + 0.2 0.6 k 0.1
0.10 0.20 0.20 0.20 0.02 0.04 0.30 0.90 0.30 0.60
iii; h o-3 0 = no 1 = yes
o-3
1 = no, 2 = yes
mm mm 1 = no, 2 = yes o-3 o-3
the
standardized
the
higher
the
listed
2 SE. Asterisk
in
Table
discriminant
discriminant coefficient
(P
23.3% 44.20, 1.0 + 0.2 62.8%
yr
O-2
of each
Not healed (n = 40)
0.08 0.50 0.60 0.70 0.60 0.50 0.80 0.30
o-3
are
1 study
contribution
Healed (n = 77)
35.4 + 2.7 45% a.9 k 1.9 28.0 ‘- 6.8 24.2 + 7.6 2.3 +- 0.7 1.6 t 0.2 57.5%
O-2
as mean
(P
36.4 + 1.9 41.60, 10.2 ‘- 1.2 24.7 -+ 4.4 19.2 + 5.6 1.7 + 0.4 1.7 + 0.1 48.1%,
o-3
Endoscopic Anterior Posterior Floor Roof Apex Longest diameter Depth Multiple ulcer Inflammation Deformity
Not healed (n = 74)
Uncorrected 2-tail
0.03 0.02 0.40 0.35 0.03 0.70 0.20 0.70
O-2
Ds,c* Low D,,c Fasting gastrin* Postprandial gastrin*
Uncorrected 2-tail
38.5 t 2.0 3 5 1"yo 9.1 k 1.2 23.8 + 4.3 11.5 + 2.0 1.8 IL 0.4 1.7 " 0.1 48.6(%
:; 1 = yes, 2 = no
Physiological BAO* MAO* Hypersecretor
-
of Variance
31.7 2 2.4 55.8% 11.1 ? 1.7 31.7 2 7.2 39.7 t 12.7 2.1 + 0.6 1.4 + 0.2 55.8%
yr
1 = M, 2 = F
Age Body weight Blood group 0 Cigarette smokers Consumption Withdrawal Alcohol users Consumption Withdrawal Analgesic users Consumption Withdrawal Work stress Neurosis
Values
Healed (n = 43)
85. No. 2
4 wk
2 wk
Characteristics
Vol.
indicates
values
3. The
relative
was function
(ignoring
indicated
that were
by
coefficient: the
negative
0.40 0.02 0.30 0.50 0.05 0.0005 0.004 0.20 0.04 0.39
log transformed
sign),
the
analysis, effects dicated
before
greater positive
toward
the
analysis.
contribution.
coefficients
healing,
favorable
statistical
and
effects.
0.50 0.90 0.60 0.06 0.007 0.04 0.20 0.40 0.20 0.06 0.10 0.20 0.30 0.95 0.10
In
indicated negative Onset
of
the
present
unfavorable coefficients
symptoms
inafter
August
1983
PREDICTION
p
F = 4.5
01
1_
0
CIGARETTES
Figure
n=29
“=ll
Patients treated with or without cimetidine were comparable in all study characteristics. This was also true when they were compared with the corresponding cimetidine-treated and placebo-treated patients of the phase 1 study (Table 1). The number of Gelusil tablets was also not statistically different between the phase 1 and phase 2 patients, including those not receiving cimetidine (3.0 + 1.0 per day in the first 2 wk, 2.4 ? 0.6 in the second 2 wk). The predictive value of the discriminant functions derived from the phase 1 study was validated in the phase 2 study [Table 4).
age 30 yr (late onset), heavy body weight, and large ulcer diameter were identified as discriminants unfavorable for ulcer healing in patients treated with 2 wk of cimetidine. Analgesic consumption, neurosis, low DsO in the pentagastrin dose-response test, low fasting serum gastrin, and large ulcer diameter were identified as discriminants unfavorable for healing in patients treated with 4 wk of cimetidine. In patients treated with 4 wk of placebo, back pain, previous gastrointestinal bleeding, and alcohol consumption were selected as factors unfavorable for ulcer healing. Prediction of healing. The discriminant score of each patient in the 2-wk cimetidine, 4-wk cimetidine, and 4-wk placebo groups was computed by
Table
3. Characteristics Selected for Discriminant and 4 wk of Placebo Treatment
Early/late
Body weight Longest ulcer
u
1 = 530
yr
2 = >30
yr
kg mm
This study shows that in any given patient, the probability of success with one of the three treatment regimens, namely, 2-wk cimetidine, 4-wk cimetidine, and 4-wk without cimetidine, can be accurately predicted. This can be done by calculating the corresponding discriminant scores based on
of Healing
After
2 and 4 wk of Cimetidine
1.21
0.06 0.20
Selected variables
s 0.6
0.5
Analgesic consumption Neurosis
0.7
Qm:
Value o-2
TreatmeQt
4 wk placebo
4 wk cimetidine
Value onset
Discussion
Functions
z wk cimetidine Selected variables
409
Phase 2 Study
>I0
PER DAY
Relationship between cigarette consumption and ulcer diameter in 79 cimetidine-treated patients of phase 1 study. A significant linear correlation was present (r = 0.32, p < 0.01).
2
HEALING
adding together the products of the values of selected discriminants and their corresponding unstandardized coefficients, plus a constant [Table 3). Each patient was classified as healed or unhealed according to the probability of his or her score belonging to the healed or unhealed group. A patient was classified as healed if the probability of healing was at least 5O%, which corresponded to the discriminant scores below which ulcers would be classified as healed (Figures 3 and 4). The prediction of healing in all three groups was sensitive, specific, and efficient.
iL
n=39
OF DUODENAL-ULCER
LJ
s
1.87
0.7
Selected variables Back pain
Value Nil = 0
U
S
2.41
1.0
0.34
0.7
1.16
0.6
Yes = 1 o-3 ngikg.
0.40 h
-0.62
0.3 -0.4
diameter
Bleeding Alcohol
n o-2
consumption
Fasting gastrin Longest
ulcer
pmol/L mm
-1.73 0.13
-0.7 0.5
diameter Constant
-6.75
Characteristics entered for discriminant analysis are those shown in Table g/day); log transformation of the asterisked characteristics [Table 2) was each patient can be calculated as the sum of individual values multiplied constant. The smaller the score, the better it is for healing. The relative standardized coefficients (S).
2.20
-1.41
2 and include also the dose of cimetidine (1 = 0.5 g/day, 2 = 1 performed before discriminant analysis. Discriminant score of by the corresponding unstandardized coefficients (U) plus the contribution of each characteristic can be compared using the
410
LAM AND
GASTKOENTEKOLOCY
KOO
.
.
.
.
-4
-6
.
.
-2
1
2
b ’ ;
- I
3
’ ;,
:--1( WK PLACEBO
I :
:I
.
: . .
’ : 3, 0 *I/%0 .yJAa
_;
.
.
_j
.
-i
0
1
2
3
DlSCRlMlNANT SCORE Figure
3. Duodenal ulcer healing and discriminant score in 119 cimetidine-treated and 58 non-cimetidine-treated patients. The solid lines indicate the scores that separate healed (solid circles and triangles) from unhealed (open circles and triangles) ulcers. The area within the dotted lines represents the indeterminate zone. Circles in phase 2 denote patients in phase 1 study, triangles study.
Phase
1
0 0
85. No.
healed unhealed
Phase
2
. II
healed unhealed
P
0
0,
??
.
. ?? ? ?
x 0
063 0 16 A 33 A 7
,_
A 21 A 19
??14
t
o 24 . 6
4 m
A 14
A
B
w
2
WK
4
CIMETIDINE
WK
4
0
CIMETIDINE
b0
CISCRIMINANT
Figure
4. Probability
of healing
2
a handful of clinical, personal, physiologic, and endoscopic characteristics. In practice, such calculations can be simply accomplished with a desk calculator. The likelihood of success with any of the treatment forms can be further improved if only patients whose scores fall outside the indeterminate zone (Figure 3) or whose scores pertain to a high probability of healing, as indicated on the probability curves (Figure 4), are selected for a particular form of treatment. Thus, up to 30% of our patients with duodenal ulcer may be selected for treatment without cimetidine, 50% for treatment with cimetidine, half of whom may be treated for only 2 wk, and the remaining 20% may be tried on another regimen of treatment. This should help to cut down the cimetidine bill substantially. The identification of the discriminant factors for healing in the present study also lends insight into the pathophysiology of duodenal ulcer healing. Past attempts to identify the healing factors can be criticized as not composite and comprehensive in approach, and only a few factors have become apparent. The adverse effect of cigarette smoking is generally accepted (1525-29); acid secretion (3,15,24,29,30), sex (15,28), and age (15,28) have been controversial; and alcohol appears favorable (28). Thus, out of 40 prospectively obtained clinical, personal, physiologic, and endoscopic characteristics in this study, individual analysis using stringent
2 WK CIMETIDINE
0
Vol.
of duodenal
WK
PLACEBO
A
%A
SCORE
ulcer
based
on discriminant
score.
August
PREDICTION
1983
Table 4. Prediction of Duodenal Ulcer Healing After Treatment With 2-wk Cimetidine, 4-wk Cimetidine, and 4-wk Placebo
Number Phase 1 Sensitivity Specificity Efficiency Positive predictive value Negative predictive value Phase 2 Healed Unhealed
4-wk placebo
4-wk cimetidine
2-wk cimetidine
%
Number
%
Number
%
29139 36140 65179 29/33
74.4 90.0 82.3 87.9
63163 14116 77179 63165
100 87.5 97.5 96.9
12114 20124 32138 12116
85.7 83.3 84.2 75.0
39149
79.6
14114
100
20122
90.9
19121 17119
90.5 89.5
33133 617
100 83.3
516 12114
83.3 85.7
statistical criteria identifies only two pertinent factors: cigarette smoking, including the quantity of cigarettes consumed, and ulcer size, both diameter and depth. In the discriminant analysis of this study, cimetidine-treated patients and placebo-treated patients were analyzed separately, since factors that influence healing in the presence or absence of acid inhibition are likely to be different. Indeed, two acidrelated physiologic measurements were identified to be of discriminant value in cimetidine-treated patients; these were (1) fasting serum gastrin concentration and (2) DsO of the pentagastrin dose-response test, which measures the parietal cell sensitivity to pentagastrin. Interestingly, the lower the fasting gastrin concentration and the more sensitive the patient is to pentagastrin, the smaller is the chance of healing with cimetidine treatment. A heavy body weight appeared unfavorable for healing, and body weight is known to correlate closely with maximal acid output (31-33), although it is not known if it also correlates with other characteristics. The discriminant analysis identified ulcer diameter to be of discriminant value in healing by cimetidine, but did not select cigarette smoking as a discriminant, although analgesic consumption was regarded as unfavorable to healing. It is possible that cigarette smoking may be related in an unknown manner to one or more of the other factors selected so that in their presence it offered no additional discriminant value. Indeed, of great interest is the finding that a demonstrable correlation between cigarette consumption and ulcer diameter does exist (Figure 2). Other factors include neurosis and late-onset disease. The latter has been reported to be more group 0 predominant, less familial, and associated with more ciga-
OF DUODENAL-ULCER
HEALING
411
rette smoking as compared with early-onset duodenal ulcer (5,16,34). In the placebo-treated group, back pain, previous gastrointestinal bleeding, and alcohol consumption were selected as factors unfavorable to healing. Because of our ignorance concerning the pathophysiology of ulcer formation and healing, the interrelationship of all of these factors as unraveled by the discriminant analysis remains to be defined. It is important to stress that we have applied a stringent criterion in the discriminant analysis, requiring the system to select only those factors that have a significant F ratio for entrance into or removal from the stepwise analysis. This minimizes any bias in the final discriminant function for prediction and ensures accurate prediction in subsequent new patients. If this criterion is relaxed, more factors may be identified, but their significance may be doubtful (24). The present study also indicates that 0.5 g/day and 1 g/day cimetidine were similarly effective for the healing of duodenal ulcer in Chinese. One possibility is that the standard dose, which is established in whites, is excessive even for whites. This is possible since 0.8 g/day cimetidine was reported to be effective (35), although this was not totally established (36). Another explanation is that the Chinese have smaller parietal cell mass and acid secretory capacity than whites (12-141, and, therefore, require less cimetidine to achieve adequate acid inhibition. It is tempting to hypothesize that in patients suitable for cimetidine treatment, the dose of cimetidine required for successful healing may be less in patients with small acid secretion. Further studies are required for substantiation. Because of the heterogeneity of duodenal ulcer disease and the marked geographic variation of placebo healing as previously indicated, the discriminant factors identified in this study may not fully apply to populations outside Hong Kong, and individual centers may need to establish their own criteria for prediction. As many medical regimens, both short-term and long-term, as well as many surgical modalities of treatment have now become available, this approach by discriminant analysis should carry the potential of selecting the right form of treatment for a given individual patient.
References Winship DH. Cimetidine in the treatment of duodenal ulcer. Review and commentary. Gastroenterology 1978;74:402-6. Misiewicz JJ. Histamine HZ-receptor antagonists in short- and long-term treatment of duodenal ulcer. In: Holtermuller KH, Malagelada JR, eds. Advances in ulcer disease. Amsterdam, Oxford, Princeton: Excerpta Medica, 1980:318-29. Binder HJ, Cocco A, Crossley RJ, et al. Cimetidine in the
412
4.
5. 6. 7.
8.
9. 10. 11.
12.
13.
14.
15.
16. 17.
18.
19.
20.
LAM AND
KOO
treatment of duodenal ulcer. A multicenter double blind study. Gastroenterology 1978;74:380-7. Lam SK, Sircus W. Studies on duodenal ulcer. I. The clinical evidence for the existence of two populations. Q J Med 1975;44:369-87. Lam SK, Ong GB. Duodenal ulcers: early and late onset. Gut 1976;17:169-79. Rotter JI. Gastric and duodenal ulcer are each many different diseases. Dig Dis Sci 1981;26:154-60. Rotter JI, Sones JQ, Samloff IM, et al. Duodenal-ulcer disease associated with elevated serum pepsinogen I: an inherited autosomal dominant disorder. N Engl J Med 1979;300:63-6. Rotter JI, Petersen G, Samloff M, et al. Genetic heterogeneity of hyperpepsinogenemic I and normopepsinogenemic I duodenal ulcer disease. Ann Intern Med 1979;91:372-7. Lam SK, Ong GB. Identification of two subgroups of familial early-onset duodenal ulcers. Ann Intern Med 1980;93:540-4. Grossman MI. Abnormalities of acid secretion in patients with duodenal ulcer. Gastroenterology 1978;75:524-6. Lam SK. Physiologic abnormalities and heterogeneity in peptic ulcer. In: Rotter JI, Samloff IM, Rimoin DL, eds. The genetics and heterogeneity of common gastrointestinal disorders. New York, London, Toronto, Sydney, San Francisco: Academic Press, 1980:67-80. Fung WP. Gastric acid secretion in Chinese with particular reference to the dose of histamine required for maximal stimulation. Gut 1970;11:955-61. Cheng FCY, Lam SK, Ong GB. Maximal acid output to graded doses of pentagastrin and its relation to parietal cell mass in Chinese patients with duodenal ulcer. Gut 1977;18:827-32. Lam SK, Hasan M, Sircus W, Wong J, Ong GB, Prescott RJ. Comparison of maximal acid output and gastrin response to meals in Chinese and Scottish normal and duodenal ulcer subjects. Gut 1980;21:324-8. Lam SK, Lam KC, Lai CL, Yeung CK, Yam LYC, Wong WS. Treatment of duodenal ulcer with antacid and sulpiride. A double-blind controlled study. Gastroenterology 1979; 76:315-22. Jirasek V. Hereditary factors in the aetiology of peptic ulcer. Acta Univ Carol Med (Praha) 1971;17:383-455, Lam SK, Ong GB. Relationship of postprandial serum gastrin response to sex, body weight, blood group status, familial dyspepsia, duration, and age of onset of ulcer symptoms in duodenal ulcer. Gut 1980;26:528-32 Lam SK, Lai CL. Inhibition of sulpiride on the cephalic phase of gastric acid and gastrin secretion in duodenal ulcer patients. Stand J Gastroenterol 1976;11:27-31. Lam SK, Isenberg JI, Grossman MI, Lane WH, Walsh JH. Gastric acid secretion is abnormally sensitive to endogenous gastrin released after peptone test meals in duodenal ulcer patients. J Clin Invest 1980;65:555-62. Isenberg JI, Grossman MI, Maxwell V, Walsh JH. Increased sensitivity to stimulation of acid secretion by pentagastrin in duodenal ulcer. J Clin Invest 1975;55:330-7.
GASTKOENTEROLOGY
Vol.
85, No. 2
21. Fleiss JL. Statistical methods for rates and proportions. New York: Wiley, 1973. 22. Snedecor GW, Cochran WG. Statistical methods. 6th ed. Ames, Iowa: Iowa State University Press, 1967. 23. Nie WH, Hull CH, Jenkins CG, Steinbrenner K, Bent DH. Statistical package of the social sciences. New York: McGrawHill, 1975:434-67. 24. Hill MA. Annotated computer output for regression analysis. BMDP technical report No. 48, BMDP2R and BMDPSR computer programs supplement, 1979. 25. Peterson WL, Sturdevant RAL, Frank1 DH, et al. Healing of duodenal ulcer with an antacid regimen. N Engl J Med 1977;297:341-5. 26. Harrison A, Elashoff J, Grossman MI. Cigarette smoking and ulcer disease. In: Smoking and health. A report of the U.S. Surgeon General. Hyattsville, Maryland: National Center for Health Statistics, 1979; DHEW publication No. 79-50066,9.39.21. 27. Korman MG, Shaw RG, Hansky J, Schmidt GT, Stern AI. Influence of smoking on healing rate of duodenal ulcer in response to cimetidine or high-dose antacid. Gastroenterology 1981;80:1451-3. 28. Sonnenberg A, Muller-Lissner SA, Vogel E, et al. Predictors of duodenal ulcer healing and relapse. Gastroenterology 1981; 81:1061-7. 29. Massarrat S, Eisenmann A. Factors affecting the healing rate of duodenal and pyloric ulcers with low-dose antacid treatment. Gut 1981;22:97-102. 30. Ippoliti AF, Sturdevant RAL, Isenberg JI, et al. Cimetidine versus intensive antacid therapy for duodenal ulcer. A multicenter trial. Gastroenterology 1978;74:393-5. 31. Sircus W. Gastric secretion in peptic ulcer disease with special reference to the influence of body weight, duration of disease and stenosis. In: Semb LS, Myren J, eds. Physiology of gastric secretion. Oslo: Universitetsforlaget, 1968:581-91. 32. Hobsley M, Whitfield PF, Faber RG, Parkin JV. Hypersecretion and length of history in duodenal ulceration. Lancet 1975;ii:lOl-4. 33. Prescott RJ, Lam SK, Hasan M, Sircus W, Wong J, Ong GB. Relationship between indices of body stature and gastric acid secretion in normal controls and duodenal ulcer subjects in two ethnic groups. Ital J Gastroenterol 1980;12:167-70. 34. Lam SK, Koo J, Sircus W. Early and late onset duodenal ulcers in Chinese and Scats. Stand J Gastroenterol 1983 (in press). 35. Kerr GD. Cimetidine: twice daily administration in duodenal ulcer-results of a UK and Ireland multicentre study. In: Baron JH, ed. Cimetidine in the 80s. Edinburgh: Churchill Livingstone, 1981:9-13. 36. Peden NR, Boyd EJS, Saunders JHB, Wormsley KG. Ranitidine in the treatment of duodenal ulceration. Stand J Gastroenterol 1981;16:325-9.
1983:85:403-12
Accurate Prediction of Duodenal-Ulcer Healing Rate by Discriminant Analysis SHIU-KUM
LAM
and JARLEY
KOO
Combined Gastrointestinal Unit, Departments Kong, Queen Mary Hospital, Hong Kong
Previous studies have shown that -50% and 70% of duodenal ulcers heal after 2 and 4 wk, respectively, of cimetidine, and at least one-third heal after 4 wk of placebo. In order to identify these groups of ulcers before treatment, a two-phase study was pe$ormed, including an initial double-blind trial of cimetidine vs. placebo in 120 patients, and a subsequent open study with identical protocol of cimetidine vs. no cimetidine in another 60 patients. Forty clinical, personal, physiologic, and endoscopic characteristics were prospectively obtained in each patient, and were analyzed by stepwise discriminant analysis at the end of phase 1. This identified the discriminants against healing after 2 wk of cimetidine as late onset disease, body weight, and ulcer diameter; those after 4 wk of cimetidine as analgesic consumption, neurosis, low fasting serum gastrin, low pentagastrin D 5o and ulcer diameter; and those after 4 wk of placebo as back pain, bleeding, and alcohol consumption. Based on the discriminant scores derived, the sensitivity, specificity, and eficiency of prediction for complete healing as determined endoscopically were 74.4% 90% and 82.3% for 2-wk cimetidine, lOOoh, 87.5% and 97.5% for 4wk cimetidine, and 85.7% 83.3% and 84.2% for 4wk placebo treatment. In phase 2, correct predictions were made in 36 of 40, 39 of 40, and 17 of 20 patients treated, respectively, for 2 and 4 wk with cimetidine, and 4 wk without cimetidine. Accurate Received July 19, 1982. Accepted March 3, 1983. Address requests for reprints to: Dr. S.-K. Lam, M.D., Department of Medicine, University of Hong Kong, Queen Mary Hospital, Hong Kong. This work was supported in part by University Research Grants (311/030/8009/31, 335/041/9908) and Wing Lung Bank Medical Research Fund (335/041/9967), University of Hong Kong. The authors acknowledge the expert statistical advice of Janet Elashoff. They are grateful to S. F. Pang, W. H. Tang, and J. Yu of the Computer Center, University of Hong Kong, for their computer assistance; M. Chong, D. Cheung, and B. Ng for nursing assistance; P. Yip and G. H. Joe for technical help; J. Watt for secretarial help; and Smith Kline & French Laboratories for sponsorship. 0 1983 by the American Gastroenterological Association 0016-5085/83/$3.00
of Medicine
and Surgery,
University
of Hong
prediction of duodenal-ulcer healing rate with and without cimetidine is thus possible by discriminant analysis. As many medical and surgical modalities of treatment are now available, this approach should have the potential of selecting the appropriate form of treatment for a given patient. The present double-blind controlled trial comparing placebo with two doses of cimetidine, namely 0.5 g and 1 g daily, in the healing of duodenal ulcer was performed to address two unresolved questions. First, it has been found that in a geographically variable group, -33.3%-66.6% of placebo-treated patients and an average of 70% of cimetidine-treated patients with duodenal ulcers would heal after 4-6 wk of treatment with cimetidine (1,2), and >50% would heal after only 2 wk of treatment (3). If it becomes possible to accurately predict which patients will heal or not heal after 4 wk of cimetidine treatment, which will heal after only 2 wk of cimetidine treatment, and which will heal with 4 wk of placebo treatment, this will not only be of therapeutic and economic importance, but will help to elucidate the factors that affect the healing of duodenal ulcer as well as the factors that underlie its pathophysiology. Because duodenal ulcer is most probably heterogeneous in etiology as evident clinically (4-6), genetically (7-9), and pathophysiologically (lO,ll), no single factor is likely to discriminate ulcers that heal from those that do not heal with treatment. A composite approach encompassing as many factors as are known may be the answer. This study prospectively, and as comprehensively as possible, collects the factors that might pertain to healing, including clinical, personal, physiologic, and endoscopic characteristics. In phase 1, which comprised the trial itself, all of these characteristics were subjected to stepwise discriminant analysis for healing and nonhealing. In phase 2, the accuracy of prediction was tested in a new group of patients who went through the same protocol, except that they
GASTROENTEROLOGY
404 LAM AND KOO
were randomly treated with or without cimetidine. Second, compared with whites, Chinese patients with duodenal ulcer had significantly smaller parieta1 cell mass and histamine- or pentagastrin-stimulated maximal acid output (12-14). Assuming that the adequacy of acid reduction by cimetidine depends on (a] the size of the parietal cell mass and (b) the dose of cimetidine used, it is possible that smaller doses of cimetidine than those established in whites may be adequate for the healing of duodenal ulcer in Chinese. Indeed, a recent clinical trial indicated that antacid tablets with a neutralizing capacity of only 12.5mmol HCl/tablet, given in tablet form 1 and 3 h after the three main meals of the day and at bedtime, healed 76.9% duodenal ulcers compared with 33.3% using placebo (15). If a substantially smaller dose of cimetidine proves effective, it would mean that cimetidine would become economically accessible to substantially more patients in the developing Asian countries, and would also suggest that the dose of cimetidine may be reduced in patients with small acid secretion.
Methods Patients Candidates were recruited from patients attending the University Medical and Surgical Gastrointestinal Clinics, as well as patients admitted to the general medical, surgical, and gastrointestinal wards of the University Departments of Medicine and Surgery, Queen Mary Hospital, Hong Kong. This has the advantage of including a wide spectrum of patients in the study. Candidates were excluded from consideration if they had a concomitant medical problem, particularly renal diseases, cardiovascular diseases, diabetes mellitus, or chronic obstructive airways diseases, pyloric stenosis, previous gastric operations for ulcer, or an associated gastric ulcer. The protocol of the study was approved by the Committee on Higher Degrees and Research Grants, University of Hong Kong.
Clinical and Personal
Characteristics
After being accepted into the study, the patients were interviewed in detail and the clinical and personal characteristics as listed later in Table 2 were carefully recorded. Further explanation is necessary regarding the following points: early-onset patients were defined as those having the onset of ulcer symptoms at 530 yr of age, and late-onset as those having the onset at >30 yr of age. The cutoff point at age 30 yr was based on previous studies indicating that early-onset patients differed from late-onset patients, particularly in being more familial, having the same blood group distribution as the general population instead of blood group 0 predominence, as in late-onset patients (5,16), and having higher meal-stimulated gastrin response (17). Symptom period was the number of days before the study during which the patient suffered from the current pain. Pain severity, which was also assessed at
Vol. 85. No. 2
subsequent visits, was scored as follows: nil = 0; mild = 1; off-work or work-interfering = 2; and requiring immediate medical attention (e.g., emergency room services) = 3. Pain duration was the average duration of the current pain in hours. Familial dyspepsia was considered present when the patient indicated that ulcer dyspepsia, gastrointestinal bleeding, perforation, or gastric surgery, or a combination of any or all, had occurred in a first-degree relative. This was scored as nil = 0, one first-degree relative = 1, two first-degree relatives = 2, and three or more first-degree relatives or both parents = 3. Stress at work was scored subjectively by the patient as 0-3, and included work pressure or labor intensity of the job, or both. Neurosis was scored according to the number of neurotic traits present with a maximum of 3 (insomnia, tension headache, palpitation, hand tremors, nonpostural dizzy spells). Habitual cigarette smoking was scored as nil = 0, %lO/day = 1, and >lO/day = 2. The majority of smokers had smoked continuously for >lO yr, and all smokers had this habit for at least 5 yr. Habitual alcohol consumption for at least 5 yr was scored as nil = 0, estimated alcohol ~50 g/day = 1, >50 g/day = 2. Habitual analgesic consumption for at least 1 yr was scored as nil = 0, 1 dose/m0 = 1, at least 1 doselwk = 2; analgesics included aspirin, nonsteroidal antiinflammatory agents, and the many forms of patent Chinese herbal medicine, the nature of which was mostly undefined. The patients were not told to stop their smoking, alcoholic, or analgesic habits, but were told that these might be harmful to ulcer healing. At subsequent visits, these habits were assessed by direct questioning and the consumption per day was recorded. The consumption during treatment was scored at the end of the trial as nil to begin with = 0, withdrawn completely = 1, reduced by >50% = 2, and unchanged = 3. In addition, before and at each subsequent visit, the following nonulcer symptoms were recorded: diarrhea, constipation, skin rash, dizziness, headache, palpitation, blurring of vision, dryness of mouth, gynecomastia or galactorrhea, impotence, and libido (frequency of sexual intercourse over the past week). Each patient was also asked to record the occurrence of pain, its severity, and the number of antacid tablets consumed, on a tabulated diary card.
Medications Two doses of cimetidine, 1 g/day and 0.5 g/day, were chosen to examine the hypothesis that Chinese patients required a substandard dose of cimetidine. Since some of the potential discriminating factors for healing may be directly or indirectly related to acid secretion and, therefore, acid inhibition, the use of full and half doses of cimetidine may have an added advantage of bringing out the effects of these factors on healing. Cimetidine tablets, 100 mg or 200 mg, or placebo (Smith Kline & French Laboratories, Philadelphia, Pa.), all identical in appearance and taste, were packaged in bottles containing weekly supplies of 35 tablets and carrying a code number and printed instructions. There were four bottles under each code number. Each patient was given one bottle at each visit and was asked to take one tablet immediately before breakfast, lunch, and supper, and two tablets before retir-
August 1983
PREDICTIONOF DUODENAL-ULCER HEALING 405
ing at night. At each visit, each patient was also given 40 antacid tablets [Gelusil (Parke-Davis, Santurce, P.R.] containing magnesium trisilicate 500 mg and dried aluminium hydroxide gel 250 mg with a neutralizing capacity of 6 mEq of HCl], and was asked to chew one or two tablets for pain relief. Patients were instructed to eat no more than three regular meals per day, and not to take any other medicine without notifying the trial clinic sister. They were told to bring along at each visit any remaining medications, which would then be counted. Physiologic
Low DSoc previously in 100 normal Chinese subjects (14). was arbitrarily defined as
Characteristics Endoscopic
On two separate mornings before the commencement of treatment, the following two tests were performed in random order after fasting from midnight. After a blood sample was taken for fasting serum gastrin, the patient ate a standard meal of 50 g of protein, 40 g of carbohydrate, and 40 g of fat in the form of a milk drink [60 g of Sustagen (Mead Johnson & Company, Evansville, Ind.) in 60 ml of milk] and Chinese-style rice congee (130 g of minced beef in 240 ml of water and rice with 5 ml of cooking oil). Further blood samples were taken at 30, 60, and 120 min after the start of the meal. Serum from all samples was immediately extracted by centrifugation and stored at - 20°C. Gastrin estimations by radioimmunoassay were done in duplicate as previously detailed (18). Sensitivity was below 2.5pmol/L of serum. Within and between assay coefficients of variation were 6.2% and 13.7%,respectively. The antiserum reacted equally with G-17 and G-34 gastrin. The integrated gastrin response after the standard meal of each subject was calculated by computing the total area under the response curve from 0 to 120 min. On a separate day, gastric acid output in response to infusions of pentagastrin administered as follows was measured: 0 (0.15M saline, 1.5mmol/h), 93.8, 375,1500, and 6000 nglkg * h. Each dose of pentagastrin was infused for 30 min into a forearm vein by a syringe infusion pump (Harvard Apparatus Co., Inc., Millis, Mass.). The final dose, 6000 ng/kg * h, was infused for 60 min. Gastric juice was collected at 15-min intervals. The acid output during saline infusion was the basal acid output (BAO) and that was the during infusion of 6000 ng/kg - h pentagastrin maximal acid output (MAO). The dose required for halfmaximal response (DsO)was estimated by linear regression analysis according to the equation: response = calculated maximal response - DsO (response/dose), which gave Dso as the negative slope. The D5,, corrected for basal acid output (DsoC) was calculated according to the equation: D5,,c = D5,, [l - basal acid output/(calculated maximal acid output + basal acid output)]. The DsOc was taken to represent the sensitivity of the parietal cells in response to pentagastrin stimulation; the methodologies, validation experiments, and calculations were detailed previously (19,20). Based on the above, the physiologic characteristics, as listed later in Table 2, were bbiainedy A hypersecretor was defined as a patient having MAO/kg total body wt (correction for body weight reduces differences between the two sexes) above 2 SD of the mean of normal as established
Characteristics
Endoscopy was performed with a forward-viewing panendoscope (Olympus GIF-P, Olympus Corp. of America, New Hyde Park, N.Y.) within 72 h before the commencement of treatment, and at the end of 2 and 4 wk of treatment, with each patient endoscoped throughout the study by the same author who was unaware of the patient’s form of treatment and clinical progress. The diameters and depth of the ulcer were estimated using the tip of the biopsy forceps. Patients were excluded from the study if the longest diameters of their ulcers were <3 mm or >25 mm. The endoscopic characteristics, as listed later in Table 2, were recorded. Ulcer site was described as anterior, posterior, floor, roof, and apex (near the exit) of the duodenal bulb. The degree of surrounding inflammation was scored as gross hyperemia, granularity, or edema (or any combination thereof) = 3, moderate = 2, mild = 1, absent = 0. The degree of deformity of the duodenal bulb was scored: actual narrowing of entrance or exit of the duodenal bulb by scar but still admitting the endoscope (external diameter = 9 mm] = 3, bilateral deformity = 2, unilateral deformity of bulb = 1,absent = 0.The findings at the end of 2 and 4 wk of treatment were compared with those observed at the initial endoscopy, and the patient was classified immediately as healed (complete disappearance of ulcer with or without residual inflammation usually in the form of hyperemia and granularity) or unhealed (persistance of ulcer, however small). Statistical
Analysis
On the assumption that 70% of cimetidine-treated patients and 35% of placebo-treated patients would heal after 4 wk, we calculated that we should include 37 subjects in each group to show a difference of 35% with a type I error of 0.05 and a type II error of 0.15(one-sided test) (21). Trial medications were prepared so that each treatment group consisted of 40 patients. Chi-square test with Yates’ correction, one-way analysis of variance, and paired t-test were used as appropriate (22). Values were expressed as mean + SE, and p values <0.05 were considered significant. Study
Design
Phase 1. This was the double-blind controlled study itself, as previously detailed, and consisted of 4 wk of treatment with endoscopic assessment at the end of the 2nd and 4th week.
406 LAM
(;r\S'rKOEN1‘EKOI,O(;\r' L'ol. 85 No.
AND K00
Phase 2. A further series of 60 patients were recruited by the same criteria. They were randomly allocated to two forms of treatment in a proportion of 2 : 1. These were, respectively: (a) cimetidine 1 g/day (200 mg t.d.s. immediately before meals and 400 mg at bedtime) and (b) Gelusil tablets (one or two to be chewed for pain relief). The protocol was otherwise exactly the same as in phase 1.
Stepwise
Discriminant
2
sal Ine
Analysis
At the end of the phase 1 study, patients were divided into those who had received cimetidine (0.5 glday or 1 g/day) and those given placebo. Three groups of patients, namely those who had either 2-wk cimetidine, 4wk cimetidine, or 4-wk placebo, were subjected to stepwise discriminant analysis with healed ulcer as the dependent variable and the characteristics listed later in Table 2 as the independent variables, using the Statistical Package for the Social Sciences (SPSS) programs (23). The computer employed was the IBM OS/VSl.IPO (IBM, White Plains, N.Y.). The criterion used was Rao’s V. This method maximized separation of healed and unhealed groups by weighing and combining the discriminating variables in a linear form. The single best discriminant variable was first selected, the remaining potential variables were then tested, and the variable that best improved the criterion (Rao’s V) in combination with the first variable was then chosen. Subsequent variables were selected according to their contributions to further discrimination, and previously selected variables were reexamined at each step. If a variable reduced discrimination when combined with subsequently selected variables, it was removed, but may have been subsequently reentered if proven again to be of value. Variables were thus added until additional variables no longer contributed significantly (p < 0.05) to further discrimination (change in Rao’s V). The standardized and unstandardized canonical discriminant function coefficients were calculated for the selected variables. From the standardized coefficients, the relative contribution of the variables in discriminating patients with healed ulcer from those with unhealed ulcer could be realized (vide infra). From the unstandardized coefficients, the discriminant score for each patient could be calculated (vide infra), and the patients were classified based on probability into “healed” and “unhealed” membership groups. To avoid overoptimistic bias in the final discriminant function, a conditional rule was used in the stepwise selection of variables, so that their maximum F-TO-ENTER and their minimum F-TO-REMOVE carried a p value of at least 0.05 (i.e., p IN and p OUT ~0.05).This would yield the least number of variables with the best discriminant function (24). In the discriminant analysis for the 4-wk placebo treatment, however, no variable was selected by this stringent rule, and the p IN and p OUT values were relaxed to 0.1 for that particular analysis. The accuracy of the prediction of healing was tested by asking the computer to classify the phase 2 patients into healed and unhealed, based on the selected discriminants derived from the phase 1 study. Information concerning healing of the phase 2 patients was withheld from the computer.
1
0
93.9
1
I
375
1500
6000
6000
PENTAGASTRIN INFUSION ng.Kg-'.h-' Figure
I. Pentagastrin dose-response studies after bolus injections of saline, 100 mg of cimetidine, or 200 mg of cimetidine in 6 duodenal ulcer patients. No significant differences between 100 mg and 200 mg cimetidine. Mean t SEM.
Results Cimetidine
and
Pentagastrin
Dose Response
Significant inhibition in acid output was observed after administration of 100 mg or 200 mg cimetidine. No differences in acid output were observed after administration of 100 mg and 200 mg cimetidine (Figure 1). Mean DsUC (nanograms per hour per kilogram) on the saline day [24.4 t 5.6 (SE)] was significantly lower (p < 0.03) than that after 100 mg cimetidine [289.5 2 91.9 (SE)] and after 200 mg cimetidine (391.2 ? 118.1). There was no significant difference
between
100 mg
and
200
mg cimetidine
(p < 0.3).
Phase
1 Study
group Patients. Two patients in the placebo and one in the 0.5 glday cimetidine group defaulted. One in each became asymptomatic and refused subsequent endoscopies, and the other left to attend to urgent business in China. No significant difference in each of the 40 patient characteristics listed in Table 2 could be detected among patients in the placebo group, 0.5 g/day cimetidine group, and 1 g/day cimetidine group (F ratio ranged between 0.01 and 2.3,x2 between 0.4 and 1.1, p values between 0.9 and 0.1). Some of these characteristics are presented in Table 1. The mean number of “cimetidine” tablets omitted per week in the placebo group, 0.5 g/day cimetidine group, and 1 g/day cimetidine group were 2.0 + 0.5, 2.0 -+ 0.6, and were mean
2.3 ? 0.7, respectively,
not significantly different number of Gelusil tablets
and
from each other. The taken per day in the
first 2 wk was not significantly different among these treatment groups (1.9 * 0.4, 2.0 ? 0.5, and 2.8 -t 0.9,
August 1983
Table
PREDICTION OF DUODENAL-ULCER HEALING
1. Comparison of Pretrial Daily Cimetidine, and
Treated Characteristics As WeI1 As Healing Rates of Patients 1 g Daily Cimetidine in the Phase 1 and Phase 2 Studies
With Placebo,
Phase 1 Placebo Total No. of patients Age (yr, range) Male Duration of symptom (yr) Familial dyspepsia, positive n scorea Blood group 0 No. with previous bleeding Cigarette smokers Alcohol users Analgesic users BAO (mmolih) MAO (mmol/h) DX (ng/kg * h) Fasting gastrin CfmoI/ml) Postprandial gastrin (nmol.min/L) Ulcerdiameter(mm) Ulcer depth (mm) Ulcer healing % healed z wk vs. placebo (p) % healed 4 wk vs. placebo (p)
Cimetidine (0.5g/day)
407
0.5 g
Phase z Cimetidine (1 g/day)
P
Cimetidine (1 g/day1
No cimetidine
38 44.0(16-75) 28 9.22 1.9 22 1.72 0.1 16 22 21 6 5 3.6f 0.5 25.5-+1.3 139.1k 26.0 36.7t 5.8 9.4k 1.0
47.2(19-75) 29 8.5* 1.3 19 1.9f 0.2 17 26 20 8 4 2.72 0.3 25.3-t1.7 129.9* 19.8 39.0? 4.9 10.6? 1.8
40 48.6(19-73) 28 11.5+ 1.9 23 1.82 0.2 24 23 19 5 6 2.5k 0.3 25.3k 1.6 168.4+ 29.1 32.02 8.3 7.8+ 0.5
CO.5 Cl.0
40 41.1(20-68) 27 9.1" 1.6 22 1.8f 0.2 21 23 19 6 6 3.0+ 0.4 24.8k 1.7 143.1+ 22.5 40.3+ 3.8 9.8" 1.2
20 43.5(25-64) 15 11.12 2.4 9 1.8f 0.1 11 12 10 3 3 2.8f 0.2 26.2f 1.3 140.42 23.0 36.42 3.9 8.2k 1.1
7.42 0.6 2.1+ 0.1
7.1+ 0.6 2.2t 0.1
7.2t 0.6 2 t 0.1
CO.9 CO.5
6.92 0.7 1.9k 0.1
7.25 0.6 2.0* 0.1
10.5
48.7 <0.0005 82.1 <0.0005
52.5
10.0
<0.0005 77.5 <0.0005
82.5
30.0
36.8
39
50.0
BAO = basal acid output. MAO = maximal acid output. D5~ = corrected dose required for half-maximal response. No differences in healing rates between the two cimetidine groups were observed at 2 wk (p < 0.9) and 4 wk (p < 0.5). No differences among the three cimetidine groups and between the placebo and no cimetidine group were observed. Mean 2 SEM. ’ See Methods section.
respectively), but were significantly (p < 0.02) more in the placebo (2.2 + 0.5) than in the 0.5 g/day cimetidine (0.5 + 0.2) or in the 1 g/day cimetidine (0.5 + 0.3) group in the second 2 wk. No gynecomastia, galactorrhea, impotence, change in libido, or hematologic or biochemical abnormality was recorded in any patient. Minor nonulcer symptoms (vide supra) were noted as new events in 15.8%, 15.4%, and lo%, as old complaints in 26.3%, 30.7%, and 30%, and to be better in 10.5%, 10.3%, and 12.5% of patients treated, respectively, with placebo, 0.5 g/day cimetidine, and 1 g/day cimetidine. There were no significant differences among groups. Healing. Both 0.5 g/day and 1 g/day cimetidine healed significantly more ulcers than placebo after 2 or 4 wk of treatment, and their healing rates were not significantly different from each other. In relation to these findings, it is pertinent to note that the corresponding healing rates of patients treated with and without cimetidine in the phase 2 study were similar to those observed in phase 1 (Table 1). After 2 and 4 wk of treatment, 29.3% and 7.3% of patients with healed ulcer, respectively, remained symptomatic, whereas 75% and 74.3% of patients with unhealed ulcer, respectively, were asymptomatic.
Individual analysis. Compared with those who healed, patients who did not heal after 2 wk of treatment were significantly older at the onset of symptoms, and had a shorter symptom period and greater body weight. More of them were cigarette smokers who smoked more heavily and who abstained less during treatment. More of them were analgesic users. More had greater ulcer diameter and depth, less posterior and more apical ulceration, more multiple ulceration, and more inflammation. The group that did not heal at the end of 4 wk contained significantly more cigarette smokers who smoked more heavily and significantly more subjects with greater ulcer diameter and more apical ulceration. A greater proportion in this group had low DSO in the pentagastrin dose-response test. Because multiple comparisons were made (n = 40), the p values were corrected by multiplying by a factor of 40. When this was done, only cigarette smoking (p < 0.002), number of cigarettes smoked (p < 0.011, ulcer diameter (p < O.OOZ), and ulcer depth (p < 0.02) remained significant (Table 2). A significant correlation was observed between cigarette consumption and ulcer diameter (Figure 2). Discriminant analysis. The selected discriminant for healing as based on patients in the
408
LAM AND
Table
2.
CASTROENTEROLOCY
KOO
Contribution
of Various
Characteristics
to Healing
as Analyzed
Individually
by x2 or Analysis
Individual values
Clinical Onset age Early/late onset, early onset Duration of symptoms Remission period Symptom period Pain duration Pain score Nocturnal awakening pain Back pain Familial dyspepsia, positive strength Bleeding, positive Personal Sex
phase
are expressed
29.7% 60.8% 0.9 k 0.1 59.5"/,
0.70 0.30 0.60 0.90
24.7”A 50.6% 0.9 2 0.1 62.30,
32.5% 62.5% 1.1 + 0.2 57.5%
0.15 0.10 0.30 0.80
72% 42.8 2 2.6 50.7 2 1.3 51.2"/, 30.2"/, 0.6 + 0.1 0.8 + 0.2 14.0% 0.1 + 0.1 0.2 + 0.1
o-3 o-3 o-3
4.701 0.1 + 0 0.3 k 0 1.4 + 0.2 0.7 + 0.1
73% 48.8 I? 1.9 54.1 5 0.9 47.3"/, 63.50, 1.1 2 0.3 1.4 + 0.2 17.60, 0.3 + 0.1 0.2 2 0.1 17.6"h 0.2 k 0.1 0.3 ? 0.1 1.1 k 0.1 0.7 lr 0.1
0.70 0.06 0.04 0.80 0.0005 0.003 0.02 0.80 0.04 0.30 0.03 0.03 0.30 0.20 0.90
70.1% 46.8 ‘- 1.8 52.7 + 0.9 54.5% 42.8”/, 0.8 + 0.1 1.1 t 0.2 13.0% 0.2 + 0.1 0.2 t 0 9.1% 0.1 + 0 0.3 2 0.1 1.2 + 0.1 0.6 + 0.1
77.5% 46.4 + 2.9 53.6 + 1.4 37.5% 67.501 1.2 k 0.1 1.4 + 0.2 22.5”Aj 0.4 -t 0.1 0.2 + 0.1 200, 0.3 -+ 0.1 0.3 2 0 1.2 2 0.2 0.9 -’ 0.2
mmolih mmol/h 1 = no, 2 = yes ng/kg . h 1 = no, 2 = yes pmol/L nmol * min/L
2.8 + 0.4 25.4 " 1.6 48.8% 158.1 2 22.5 46.S"h 31.6 k 2.7 9.7 t 0.9
3.0 k 0.3 25.2 5 1.1 62.2% 138.5 + 18.7 60.8% 38.3 + 3.7 9.0 + 1.0
0.50 0.70 0.20 0.80 0.10 0.30 0.55
2.8 + 0.3 25.7 + 1.2 55.8”/, 162.2 -+ 17.8 46.8% 37.8 + 3.1 9.4 + 1.0
3.2 k 0.4 24.9 k 1.5 6001 116.0 + 26.0 72.5% 32.2 k 5.3 9.0 Zk 1.0
0.40 0.65 0.50 0.20 0.04 0.60 0.70
48.8% 20.9% 16.2% 4.9"y" 2.3% 5.5 k 0.3 1.8 k 0.1 7.0% 1.2 2 0.1 0.7 2 0.1
41.9% 6.801 27% 10.8”/, 16.2”r a.2 ? 0.5 2.3 k 0.1 16.2% 1.5 + 0.1 0.6 + 0.1
49.3% 14.301 18.2% 10.4% 5.2% 6.6 k 0.4 2.0 2 0.1 12.3”/r, 1.5 +- 0.1 0.6 -c 0.1
35% 7.5% 32.501 5% 22.5% 8.5 ” 0.7 2.3 2 0.1 13.9”/, 1.3 + 0.2 0.6 k 0.1
0.10 0.20 0.20 0.20 0.02 0.04 0.30 0.90 0.30 0.60
iii; h o-3 0 = no 1 = yes
o-3
1 = no, 2 = yes
mm mm 1 = no, 2 = yes o-3 o-3
the
standardized
the
higher
the
listed
2 SE. Asterisk
in
Table
discriminant
discriminant coefficient
(P
23.3% 44.20, 1.0 + 0.2 62.8%
yr
O-2
of each
Not healed (n = 40)
0.08 0.50 0.60 0.70 0.60 0.50 0.80 0.30
o-3
are
1 study
contribution
Healed (n = 77)
35.4 + 2.7 45% a.9 k 1.9 28.0 ‘- 6.8 24.2 + 7.6 2.3 +- 0.7 1.6 t 0.2 57.5%
O-2
as mean
(P
36.4 + 1.9 41.60, 10.2 ‘- 1.2 24.7 -+ 4.4 19.2 + 5.6 1.7 + 0.4 1.7 + 0.1 48.1%,
o-3
Endoscopic Anterior Posterior Floor Roof Apex Longest diameter Depth Multiple ulcer Inflammation Deformity
Not healed (n = 74)
Uncorrected 2-tail
0.03 0.02 0.40 0.35 0.03 0.70 0.20 0.70
O-2
Ds,c* Low D,,c Fasting gastrin* Postprandial gastrin*
Uncorrected 2-tail
38.5 t 2.0 3 5 1"yo 9.1 k 1.2 23.8 + 4.3 11.5 + 2.0 1.8 IL 0.4 1.7 " 0.1 48.6(%
:; 1 = yes, 2 = no
Physiological BAO* MAO* Hypersecretor
-
of Variance
31.7 2 2.4 55.8% 11.1 ? 1.7 31.7 2 7.2 39.7 t 12.7 2.1 + 0.6 1.4 + 0.2 55.8%
yr
1 = M, 2 = F
Age Body weight Blood group 0 Cigarette smokers Consumption Withdrawal Alcohol users Consumption Withdrawal Analgesic users Consumption Withdrawal Work stress Neurosis
Values
Healed (n = 43)
85. No. 2
4 wk
2 wk
Characteristics
Vol.
indicates
values
3. The
relative
was function
(ignoring
indicated
that were
by
coefficient: the
negative
0.40 0.02 0.30 0.50 0.05 0.0005 0.004 0.20 0.04 0.39
log transformed
sign),
the
analysis, effects dicated
before
greater positive
toward
the
analysis.
contribution.
coefficients
healing,
favorable
statistical
and
effects.
0.50 0.90 0.60 0.06 0.007 0.04 0.20 0.40 0.20 0.06 0.10 0.20 0.30 0.95 0.10
In
indicated negative Onset
of
the
present
unfavorable coefficients
symptoms
inafter
August
1983
PREDICTION
p
F = 4.5
01
1_
0
CIGARETTES
Figure
n=29
“=ll
Patients treated with or without cimetidine were comparable in all study characteristics. This was also true when they were compared with the corresponding cimetidine-treated and placebo-treated patients of the phase 1 study (Table 1). The number of Gelusil tablets was also not statistically different between the phase 1 and phase 2 patients, including those not receiving cimetidine (3.0 + 1.0 per day in the first 2 wk, 2.4 ? 0.6 in the second 2 wk). The predictive value of the discriminant functions derived from the phase 1 study was validated in the phase 2 study [Table 4).
age 30 yr (late onset), heavy body weight, and large ulcer diameter were identified as discriminants unfavorable for ulcer healing in patients treated with 2 wk of cimetidine. Analgesic consumption, neurosis, low DsO in the pentagastrin dose-response test, low fasting serum gastrin, and large ulcer diameter were identified as discriminants unfavorable for healing in patients treated with 4 wk of cimetidine. In patients treated with 4 wk of placebo, back pain, previous gastrointestinal bleeding, and alcohol consumption were selected as factors unfavorable for ulcer healing. Prediction of healing. The discriminant score of each patient in the 2-wk cimetidine, 4-wk cimetidine, and 4-wk placebo groups was computed by
Table
3. Characteristics Selected for Discriminant and 4 wk of Placebo Treatment
Early/late
Body weight Longest ulcer
u
1 = 530
yr
2 = >30
yr
kg mm
This study shows that in any given patient, the probability of success with one of the three treatment regimens, namely, 2-wk cimetidine, 4-wk cimetidine, and 4-wk without cimetidine, can be accurately predicted. This can be done by calculating the corresponding discriminant scores based on
of Healing
After
2 and 4 wk of Cimetidine
1.21
0.06 0.20
Selected variables
s 0.6
0.5
Analgesic consumption Neurosis
0.7
Qm:
Value o-2
TreatmeQt
4 wk placebo
4 wk cimetidine
Value onset
Discussion
Functions
z wk cimetidine Selected variables
409
Phase 2 Study
>I0
PER DAY
Relationship between cigarette consumption and ulcer diameter in 79 cimetidine-treated patients of phase 1 study. A significant linear correlation was present (r = 0.32, p < 0.01).
2
HEALING
adding together the products of the values of selected discriminants and their corresponding unstandardized coefficients, plus a constant [Table 3). Each patient was classified as healed or unhealed according to the probability of his or her score belonging to the healed or unhealed group. A patient was classified as healed if the probability of healing was at least 5O%, which corresponded to the discriminant scores below which ulcers would be classified as healed (Figures 3 and 4). The prediction of healing in all three groups was sensitive, specific, and efficient.
iL
n=39
OF DUODENAL-ULCER
LJ
s
1.87
0.7
Selected variables Back pain
Value Nil = 0
U
S
2.41
1.0
0.34
0.7
1.16
0.6
Yes = 1 o-3 ngikg.
0.40 h
-0.62
0.3 -0.4
diameter
Bleeding Alcohol
n o-2
consumption
Fasting gastrin Longest
ulcer
pmol/L mm
-1.73 0.13
-0.7 0.5
diameter Constant
-6.75
Characteristics entered for discriminant analysis are those shown in Table g/day); log transformation of the asterisked characteristics [Table 2) was each patient can be calculated as the sum of individual values multiplied constant. The smaller the score, the better it is for healing. The relative standardized coefficients (S).
2.20
-1.41
2 and include also the dose of cimetidine (1 = 0.5 g/day, 2 = 1 performed before discriminant analysis. Discriminant score of by the corresponding unstandardized coefficients (U) plus the contribution of each characteristic can be compared using the
410
LAM AND
GASTKOENTEKOLOCY
KOO
.
.
.
.
-4
-6
.
.
-2
1
2
b ’ ;
- I
3
’ ;,
:--1( WK PLACEBO
I :
:I
.
: . .
’ : 3, 0 *I/%0 .yJAa
_;
.
.
_j
.
-i
0
1
2
3
DlSCRlMlNANT SCORE Figure
3. Duodenal ulcer healing and discriminant score in 119 cimetidine-treated and 58 non-cimetidine-treated patients. The solid lines indicate the scores that separate healed (solid circles and triangles) from unhealed (open circles and triangles) ulcers. The area within the dotted lines represents the indeterminate zone. Circles in phase 2 denote patients in phase 1 study, triangles study.
Phase
1
0 0
85. No.
healed unhealed
Phase
2
. II
healed unhealed
P
0
0,
??
.
. ?? ? ?
x 0
063 0 16 A 33 A 7
,_
A 21 A 19
??14
t
o 24 . 6
4 m
A 14
A
B
w
2
WK
4
CIMETIDINE
WK
4
0
CIMETIDINE
b0
CISCRIMINANT
Figure
4. Probability
of healing
2
a handful of clinical, personal, physiologic, and endoscopic characteristics. In practice, such calculations can be simply accomplished with a desk calculator. The likelihood of success with any of the treatment forms can be further improved if only patients whose scores fall outside the indeterminate zone (Figure 3) or whose scores pertain to a high probability of healing, as indicated on the probability curves (Figure 4), are selected for a particular form of treatment. Thus, up to 30% of our patients with duodenal ulcer may be selected for treatment without cimetidine, 50% for treatment with cimetidine, half of whom may be treated for only 2 wk, and the remaining 20% may be tried on another regimen of treatment. This should help to cut down the cimetidine bill substantially. The identification of the discriminant factors for healing in the present study also lends insight into the pathophysiology of duodenal ulcer healing. Past attempts to identify the healing factors can be criticized as not composite and comprehensive in approach, and only a few factors have become apparent. The adverse effect of cigarette smoking is generally accepted (1525-29); acid secretion (3,15,24,29,30), sex (15,28), and age (15,28) have been controversial; and alcohol appears favorable (28). Thus, out of 40 prospectively obtained clinical, personal, physiologic, and endoscopic characteristics in this study, individual analysis using stringent
2 WK CIMETIDINE
0
Vol.
of duodenal
WK
PLACEBO
A
%A
SCORE
ulcer
based
on discriminant
score.
August
PREDICTION
1983
Table 4. Prediction of Duodenal Ulcer Healing After Treatment With 2-wk Cimetidine, 4-wk Cimetidine, and 4-wk Placebo
Number Phase 1 Sensitivity Specificity Efficiency Positive predictive value Negative predictive value Phase 2 Healed Unhealed
4-wk placebo
4-wk cimetidine
2-wk cimetidine
%
Number
%
Number
%
29139 36140 65179 29/33
74.4 90.0 82.3 87.9
63163 14116 77179 63165
100 87.5 97.5 96.9
12114 20124 32138 12116
85.7 83.3 84.2 75.0
39149
79.6
14114
100
20122
90.9
19121 17119
90.5 89.5
33133 617
100 83.3
516 12114
83.3 85.7
statistical criteria identifies only two pertinent factors: cigarette smoking, including the quantity of cigarettes consumed, and ulcer size, both diameter and depth. In the discriminant analysis of this study, cimetidine-treated patients and placebo-treated patients were analyzed separately, since factors that influence healing in the presence or absence of acid inhibition are likely to be different. Indeed, two acidrelated physiologic measurements were identified to be of discriminant value in cimetidine-treated patients; these were (1) fasting serum gastrin concentration and (2) DsO of the pentagastrin dose-response test, which measures the parietal cell sensitivity to pentagastrin. Interestingly, the lower the fasting gastrin concentration and the more sensitive the patient is to pentagastrin, the smaller is the chance of healing with cimetidine treatment. A heavy body weight appeared unfavorable for healing, and body weight is known to correlate closely with maximal acid output (31-33), although it is not known if it also correlates with other characteristics. The discriminant analysis identified ulcer diameter to be of discriminant value in healing by cimetidine, but did not select cigarette smoking as a discriminant, although analgesic consumption was regarded as unfavorable to healing. It is possible that cigarette smoking may be related in an unknown manner to one or more of the other factors selected so that in their presence it offered no additional discriminant value. Indeed, of great interest is the finding that a demonstrable correlation between cigarette consumption and ulcer diameter does exist (Figure 2). Other factors include neurosis and late-onset disease. The latter has been reported to be more group 0 predominant, less familial, and associated with more ciga-
OF DUODENAL-ULCER
HEALING
411
rette smoking as compared with early-onset duodenal ulcer (5,16,34). In the placebo-treated group, back pain, previous gastrointestinal bleeding, and alcohol consumption were selected as factors unfavorable to healing. Because of our ignorance concerning the pathophysiology of ulcer formation and healing, the interrelationship of all of these factors as unraveled by the discriminant analysis remains to be defined. It is important to stress that we have applied a stringent criterion in the discriminant analysis, requiring the system to select only those factors that have a significant F ratio for entrance into or removal from the stepwise analysis. This minimizes any bias in the final discriminant function for prediction and ensures accurate prediction in subsequent new patients. If this criterion is relaxed, more factors may be identified, but their significance may be doubtful (24). The present study also indicates that 0.5 g/day and 1 g/day cimetidine were similarly effective for the healing of duodenal ulcer in Chinese. One possibility is that the standard dose, which is established in whites, is excessive even for whites. This is possible since 0.8 g/day cimetidine was reported to be effective (35), although this was not totally established (36). Another explanation is that the Chinese have smaller parietal cell mass and acid secretory capacity than whites (12-141, and, therefore, require less cimetidine to achieve adequate acid inhibition. It is tempting to hypothesize that in patients suitable for cimetidine treatment, the dose of cimetidine required for successful healing may be less in patients with small acid secretion. Further studies are required for substantiation. Because of the heterogeneity of duodenal ulcer disease and the marked geographic variation of placebo healing as previously indicated, the discriminant factors identified in this study may not fully apply to populations outside Hong Kong, and individual centers may need to establish their own criteria for prediction. As many medical regimens, both short-term and long-term, as well as many surgical modalities of treatment have now become available, this approach by discriminant analysis should carry the potential of selecting the right form of treatment for a given individual patient.
References Winship DH. Cimetidine in the treatment of duodenal ulcer. Review and commentary. Gastroenterology 1978;74:402-6. Misiewicz JJ. Histamine HZ-receptor antagonists in short- and long-term treatment of duodenal ulcer. In: Holtermuller KH, Malagelada JR, eds. Advances in ulcer disease. Amsterdam, Oxford, Princeton: Excerpta Medica, 1980:318-29. Binder HJ, Cocco A, Crossley RJ, et al. Cimetidine in the
412
4.
5. 6. 7.
8.
9. 10. 11.
12.
13.
14.
15.
16. 17.
18.
19.
20.
LAM AND
KOO
treatment of duodenal ulcer. A multicenter double blind study. Gastroenterology 1978;74:380-7. Lam SK, Sircus W. Studies on duodenal ulcer. I. The clinical evidence for the existence of two populations. Q J Med 1975;44:369-87. Lam SK, Ong GB. Duodenal ulcers: early and late onset. Gut 1976;17:169-79. Rotter JI. Gastric and duodenal ulcer are each many different diseases. Dig Dis Sci 1981;26:154-60. Rotter JI, Sones JQ, Samloff IM, et al. Duodenal-ulcer disease associated with elevated serum pepsinogen I: an inherited autosomal dominant disorder. N Engl J Med 1979;300:63-6. Rotter JI, Petersen G, Samloff M, et al. Genetic heterogeneity of hyperpepsinogenemic I and normopepsinogenemic I duodenal ulcer disease. Ann Intern Med 1979;91:372-7. Lam SK, Ong GB. Identification of two subgroups of familial early-onset duodenal ulcers. Ann Intern Med 1980;93:540-4. Grossman MI. Abnormalities of acid secretion in patients with duodenal ulcer. Gastroenterology 1978;75:524-6. Lam SK. Physiologic abnormalities and heterogeneity in peptic ulcer. In: Rotter JI, Samloff IM, Rimoin DL, eds. The genetics and heterogeneity of common gastrointestinal disorders. New York, London, Toronto, Sydney, San Francisco: Academic Press, 1980:67-80. Fung WP. Gastric acid secretion in Chinese with particular reference to the dose of histamine required for maximal stimulation. Gut 1970;11:955-61. Cheng FCY, Lam SK, Ong GB. Maximal acid output to graded doses of pentagastrin and its relation to parietal cell mass in Chinese patients with duodenal ulcer. Gut 1977;18:827-32. Lam SK, Hasan M, Sircus W, Wong J, Ong GB, Prescott RJ. Comparison of maximal acid output and gastrin response to meals in Chinese and Scottish normal and duodenal ulcer subjects. Gut 1980;21:324-8. Lam SK, Lam KC, Lai CL, Yeung CK, Yam LYC, Wong WS. Treatment of duodenal ulcer with antacid and sulpiride. A double-blind controlled study. Gastroenterology 1979; 76:315-22. Jirasek V. Hereditary factors in the aetiology of peptic ulcer. Acta Univ Carol Med (Praha) 1971;17:383-455, Lam SK, Ong GB. Relationship of postprandial serum gastrin response to sex, body weight, blood group status, familial dyspepsia, duration, and age of onset of ulcer symptoms in duodenal ulcer. Gut 1980;26:528-32 Lam SK, Lai CL. Inhibition of sulpiride on the cephalic phase of gastric acid and gastrin secretion in duodenal ulcer patients. Stand J Gastroenterol 1976;11:27-31. Lam SK, Isenberg JI, Grossman MI, Lane WH, Walsh JH. Gastric acid secretion is abnormally sensitive to endogenous gastrin released after peptone test meals in duodenal ulcer patients. J Clin Invest 1980;65:555-62. Isenberg JI, Grossman MI, Maxwell V, Walsh JH. Increased sensitivity to stimulation of acid secretion by pentagastrin in duodenal ulcer. J Clin Invest 1975;55:330-7.
GASTKOENTEROLOGY
Vol.
85, No. 2
21. Fleiss JL. Statistical methods for rates and proportions. New York: Wiley, 1973. 22. Snedecor GW, Cochran WG. Statistical methods. 6th ed. Ames, Iowa: Iowa State University Press, 1967. 23. Nie WH, Hull CH, Jenkins CG, Steinbrenner K, Bent DH. Statistical package of the social sciences. New York: McGrawHill, 1975:434-67. 24. Hill MA. Annotated computer output for regression analysis. BMDP technical report No. 48, BMDP2R and BMDPSR computer programs supplement, 1979. 25. Peterson WL, Sturdevant RAL, Frank1 DH, et al. Healing of duodenal ulcer with an antacid regimen. N Engl J Med 1977;297:341-5. 26. Harrison A, Elashoff J, Grossman MI. Cigarette smoking and ulcer disease. In: Smoking and health. A report of the U.S. Surgeon General. Hyattsville, Maryland: National Center for Health Statistics, 1979; DHEW publication No. 79-50066,9.39.21. 27. Korman MG, Shaw RG, Hansky J, Schmidt GT, Stern AI. Influence of smoking on healing rate of duodenal ulcer in response to cimetidine or high-dose antacid. Gastroenterology 1981;80:1451-3. 28. Sonnenberg A, Muller-Lissner SA, Vogel E, et al. Predictors of duodenal ulcer healing and relapse. Gastroenterology 1981; 81:1061-7. 29. Massarrat S, Eisenmann A. Factors affecting the healing rate of duodenal and pyloric ulcers with low-dose antacid treatment. Gut 1981;22:97-102. 30. Ippoliti AF, Sturdevant RAL, Isenberg JI, et al. Cimetidine versus intensive antacid therapy for duodenal ulcer. A multicenter trial. Gastroenterology 1978;74:393-5. 31. Sircus W. Gastric secretion in peptic ulcer disease with special reference to the influence of body weight, duration of disease and stenosis. In: Semb LS, Myren J, eds. Physiology of gastric secretion. Oslo: Universitetsforlaget, 1968:581-91. 32. Hobsley M, Whitfield PF, Faber RG, Parkin JV. Hypersecretion and length of history in duodenal ulceration. Lancet 1975;ii:lOl-4. 33. Prescott RJ, Lam SK, Hasan M, Sircus W, Wong J, Ong GB. Relationship between indices of body stature and gastric acid secretion in normal controls and duodenal ulcer subjects in two ethnic groups. Ital J Gastroenterol 1980;12:167-70. 34. Lam SK, Koo J, Sircus W. Early and late onset duodenal ulcers in Chinese and Scats. Stand J Gastroenterol 1983 (in press). 35. Kerr GD. Cimetidine: twice daily administration in duodenal ulcer-results of a UK and Ireland multicentre study. In: Baron JH, ed. Cimetidine in the 80s. Edinburgh: Churchill Livingstone, 1981:9-13. 36. Peden NR, Boyd EJS, Saunders JHB, Wormsley KG. Ranitidine in the treatment of duodenal ulceration. Stand J Gastroenterol 1981;16:325-9.