The search for meaning—Symptoms and transvaginal sonography screening for ovarian cancer

Predicting malignancy

Abstract

BACKGROUND:

The mortality rate of ovarian cancer is greater than that of all other major gynecologic malignancies. Detecting ovarian cancer at an early and curable stage long has been an objective of oncologists. Recently, it was reported that certain symptom patterns are informative for the presence of ovarian malignancy. In this article, the authors report on how symptoms and ultrasound predict ovarian malignancy.

METHODS:

Two hundred seventy-two women who were participating in annual transvaginal sonography (TVS) screening were selected from among 31,748 women who were enrolled. Symptom results were correlated with ultrasound and surgical pathology findings.

RESULTS:

TVS performed better than symptoms analysis for detecting malignancies (sensitivity, 73.3% vs 20%), and symptoms analysis performed better for distinguishing benign tumors (specificity, 91.3% vs 74.4%). The use of TVS and symptoms analysis in series resulted in poorer identification of malignancy (sensitivity, 16.7%) but improved the ability to distinguish benign tumors (specificity, 97.9%). Decisions using either symptoms or TVS combined in parallel had small increases in sensitivity (+3.3%) and had coordinated, small decreases in specificity (−5.8%).

CONCLUSIONS:

Symptoms did identify ovarian malignancies, but not as well as TVS. The current findings indicated that: 1) tumors that are negative by both ultrasound and a symptoms index are likely to be benign (specificity, >97%), and 2) adding symptoms information that has weight equal to the weight of ultrasound only slightly improves the discrimination of malignancy (sensitivity increase, +3.3%). Thus, a major benefit in discriminating malignancy was achieved through ultrasound, whereas the absence of symptoms in conjunction with an abnormal ultrasound (characterized by a low morphology index) indicated that the mass was benign and that surgery may not be required. Finally, informative symptoms can be expected to be absent in 80% of patients with ovarian malignancies. Cancer 2009. © 2009 American Cancer Society.

Ovarian cancer remains the fourth leading cause of cancer death in US women.1 This year, the lives of over 15,000 women will be claimed by ovarian cancer in the United States alone.2 Pelvic examination is notably inaccurate in detecting subtle changes in ovarian size and morphology, particularly in postmenopausal women.3 Although ovarian cancer has been perceived as a “silent killer” that produces few specific symptoms, recent studies have indicated that certain symptoms are significantly more common in women with ovarian cancer than in women in the general population.4, 5 These findings have resulted in the design of a symptoms index6 that reportedly has utility for identifying early stage ovarian cancer, which often is curable by conventional therapy.7, 8 It appears, then, that the symptoms index could have utility in a screening context to detect ovarian cancer at an earlier and more curable stage.9-17

Since 1987, the University of Kentucky Ovarian Cancer Screening Project has examined the efficacy of annual transvaginal sonography (TVS) as a screening method for ovarian cancer, providing free screening to over 31,000 women who participate in this program. The results of this screening study have been reported recently.18 The current report applies the symptoms index analysis6 to this large study group as a validation study to determine whether its addition improves the performance of ovarian cancer screening.

MATERIALS AND METHODS

Patients enrolled in the University of Kentucky Ovarian Cancer Screening Project from January 1987 to June 2008 composed the source group (n = 31,748). Approval was received from the University of Kentucky Institutional Review Board. Eligibility, exclusions, instrumentation, protocol, criteria for designating an abnormality, and data collected were as recently reported.18 Participants in the screening program received free annual screening after a normal result (ie, 74.2% were scheduled 11-13 months after a normal screen, and 7.8% were scheduled even earlier, whereas 17.9% were scheduled later). After an abnormal screening result, free repeat screening was scheduled at 4 to 6 weeks. Criteria for abnormality included ovarian volume (values >2 standard deviations above the mean volume of normal ovaries in premenopausal and postmenopausal women)19 and any cystic ovarian tumor with a solid or papillary projection into its lumen.20 Morphology indexing has been useful in predicting the risk of malignancy21-24 and was performed according to the classification of Ueland et al.20 Each tumor was given a score of from 1 to 10 according to increasing morphologic complexity and volume, as outlined in Figure 1, with increasing numerical scores for septa, papillary projections, solid areas, and extratumoral free fluid. Each final morphology index (MI) score is the sum of the volume score (1-5) and the structure score (1-5).

Figure 1. Morphology index schematic is shown.

Patients with ovarian cancer on frozen section or patients with obvious metastatic disease at laparoscopy underwent immediate exploratory laparotomy and staging. Tumors were classified histologically according to the World Health Organization system and were staged according to the International Federation of Gynecology and Obstetrics system.

The selection group consisted of those members of the source group who underwent surgery after an abnormal TVS finding (n = 450) and included women with either malignant or benign tumors. The study group was formed from the selection group and was made up of women in the selection group who had returned confident responses on the symptoms index questionnaire (n = 272; 60.4%). Excluded from this group were 178 women who had died (1 stage II and 3 stage III ovarian tumors), had withdrawn from the study, who were unwilling to take the symptoms survey, or who were not confident in their answers.

Symptoms Index Analysis

The symptoms survey in the form published by Goff et al6 was used and occupied an entire 8.5 × 11-inch page printed in landscape format. The only deviations from the symptoms survey published by Goff et al6 were that 1) dark-contrast symptoms space separators replaced the open white spaces used by Goff et al6, and 2) a single confidence assessment query was added. The dark-contrast space separators improved the readability of the form and were intended to prevent visual errors arising from row or column cross-over. The confidence query question was “How confidently did you answer these questions?” Response choices for the query were 0 = no confidence, 1 = minimally sure, 2 = more than minimally sure, 3 = pretty sure, 4 = sure, and 5 = absolutely sure of accuracy.” The response query was used to identify recall bias or responses that were affected by the respondent's memory. Thirteen individuals without malignancies were excluded because they did not have a confidence response >3, and 165 individuals were excluded because they did not wish to participate or because they could not be contacted. The criteria reported by Goff et al for any specific symptom (pelvic pain, abdominal pain, increased abdominal size, bloating, feeling full, and difficulty eating), frequency (>12 times per month), and duration (<12 months) were used as the decision rule to define an individual who was “positive for symptoms” specific to ovarian cancer.6 Cross-sectional results of the symptoms questionnaire are used here and represent a single evaluation of each participant.

Statistical Methods

Proportions were compared by using chi-square statistics or Fisher exact test. Means were compared by using t statistics. Statistical significance was determined at the .05 level. Analysis of receiver operating characteristic (ROC) curves also was performed.25-27

RESULTS

There were 31,748 women who enrolled in the University of Kentucky ovarian screening study from January 1987 to June 2008, and a family history of ovarian cancer was documented in 22.5% of these women.

Clinical Characteristics of the Women Selected for Study on the Basis of Surgery Related to TVS Screening

The candidates who were selected for the current study all were involved actively in the University of Kentucky Ovarian Screening Program and had undergone surgery because of an abnormal TVS result. The selection group contained 450 women who underwent surgery related to a TVS finding, and a family history of ovarian cancer was documented in 123 of those women (27.4%). Members of the selection group who had confident responses to the symptoms questionnaire defined the study group of 272 women. Twenty-seven percent of women in the study group reported a family history of ovarian cancer. The clinical characteristics of the patients screened are shown in Table 1. There were no significant differences between the selection group and the study group with respect to age, gravidity, parity, weight, height, body mass index (BMI), body surface area, CA 125, family history of ovarian, breast, or colon cancer, use of hormone-replacement therapy, or nulliparity (Table 1). There was no significant difference between the histopathologies of the selection group and the study group (ie, responders to the symptoms questionnaire) (Table 2).

Table 1. Clinical Characteristics of the Selection Group (n=450), the Study Group (n=272), and the Nonparticipating Group (n=178)

Variable*	Selection Group			Study Group			Nonparticipating Group
	Mean±SEM	Range	No. (%)	Mean±SEM	Range	No. (%)	Mean±SEM	Range	No. (%)
SEM indicates standard error of the mean; BMI, body mass index; BSA, body surface area; HRT, hormone-replacement therapy. * There were no significant differences in age, parity, gravidity, weight, height, BMI, BSA, or CA 125; P < .9; Student t test) or in family history, use of HRT, or nulliparity (chi-square value, 3.285; P=.511). † Significantly different (P=.052).
Age, y	58.4±0.5	32-89	450	58.4±0.6	32-89	272	58.4±0.8	34-84	178
Gravidity	2.5±0.08	0-12	450	2.6±0.1	0-12	272	2.4±0.1	0-7	178
Parity	2.2±0.07	0-10	450	2.3±0.1	0-10	272	2.1±0.1	0-6	178
Weight, pounds	163.3±1.7	80-350	450	164.3±2.2	99-350	272	161.8±2.6	80-300	178
Height, inches	64.5±0.12	54-72	450	64.5±0.2	54-72	272	64.5±0.2	57-71	178
BMI, lbs/in^2	27.6±0.3	15.1-57.8	450	27.7±0.4	18-58	272	27.4±0.4	15-51	178
BSA, m^2	1.8±0.01	1.3-2.8	450	1.8±0.01	1.4-2.8	272	1.8±0.02	1.2-2.5	178
CA 125, U/mL	37.1±11	1-1500	148	36.9±13.9	1-1550	113	37±11.9	2-279	35
Family history
Ovarian cancer			123 (27)			73 (27)			50 (28)
Breast cancer			196 (44)			116 (43)			81 (45)
Colon cancer			100 (22)			70 (26)			30 (17)†
History of HRT			145 (32)			99 (36)			55 (26)
Nulliparous			58 (13)			30 (11)			28 (16)

Table 2. Histologic Findings in the Selection Group (n=450) and the Study Group (n=272)*

Finding	No. of Patients (%)
	Selection Group	Study Group
GC/LMP, granulosa cell tumor/tumor with low malignant potential. * No significant differences were observed between the selection group versus the study group (chi-square statistic, 5.616: P=.898).
Primary ovarian cancer	49 (10.9)	30 (11)
Malignant	32 (7.1)	15 (5.5)
GC/LMP	17 (3.8)	15 (5.5)
Serous cystadenoma	192 (42.7)	114 (41.9)
Endometrioma	30 (6.7)	18 (6.6)
Mucinous cystadenoma	24 (5.7)	14 (5.1)
Cystic teratoma	21 (4.7)	12 (4.4)
Hemorrhagic cyst	9 (2)	6 (2.2)
Fibroma/thecoma/Brenner tumor	33 (7.7)	22 (8.1)
Leiomyomata	19 (4.2)	16 (5.9)
Hydrosalpinx/paratubal	29 (6.4)	16 (5.9)
Other	44 (9.4)	24 (8.8)

An individual was scored positive on the symptoms index published by Goff et al6 if any of the informative symptoms reported by Goff et al (pelvic pain, abdominal pain, increased abdominal size, bloating, feeling full, having difficulty eating) were positive for >12 days per month with a duration <12 months.6 In total, 27 of 272 women (9.9%) in the study group reported symptoms that were considered positive on the Goff symptoms index. The most common symptoms reported, but not considered informative by Goff et al, were back pain, frequent urination, urinary urgency, fatigue, and constipation (Table 3). Approximately the same number of women reported having no symptoms (n = 124; 45.6%) as reported symptoms that were not considered positive (n = 121; 44.5%) on the Goff symptoms index (Table 4). Of the 27 women who were positive on the Goff symptoms index, 6 women had ovarian cancers (4 malignancies and 2 tumors of low malignant potential) (Table 5). Thus, of the 30 women with ovarian cancer in the study group who responded to the symptoms questionnaire, 6 women (20%) satisfied the Goff et al criteria for being symptoms-positive, and 80% did not. The study group was very similar to the selection group with respect to tumor stage, and ∼75% had early stage disease (Table 6). All of the women who had true‒positive (TP) results identified by informative symptoms had early stage disease (Table 6).

Table 3. Summary of Reported Symptoms (n=272)*

Symptom	No. of Patients (%)
	Goff Positive	Goff Negative
NA indicates not applicable. * Based on the symptoms survey in the form published by Goff et al.6
Back pain	NA	85 (31.3)
Frequent urination	NA	84 (30.9)
Urinary urgency	NA	80 (29.4)
Fatigue	NA	79 (29)
Constipation	NA	74 (27.2)
Abdominal bloating	14 (5.1)	65 (23.9)
Pelvic pain	15 (5.5)	59 (21.7)
Pain with intercourse	NA	59 (21.7)
Indigestion	NA	58 (21.3)
Leg swelling	NA	58 (21.3)
Increased abdominal size	15 (5.5)	50 (18.4)
Irregular menses	NA	50 (18.4)
Diarrhea	NA	49 (18)
Difficulty breathing	NA	45 (16.5)
Bleeding after menopause	NA	33 (12.1)
Nausea/vomiting	NA	32 (11.8)
Bleeding with intercourse	NA	31 (11.4)
Palpable abdomen	NA	28 (10.3)
Feels full quickly	9 (3.3)	21 (7.7)
Weight loss	NA	21 (7.7)
Abnormal eating	3 (1.1)	14 (5.1)

Table 4. Response to Symptoms Questionnaire (n=272)

Reported Symptoms	No. of Patients (%)
Reported experiencing no symptoms	124 (45.6)
Reported symptoms not considered positive on the symptoms index	121 (44.5)
Reported symptoms considered positive on the Goff symptoms index	27 (9.9)

Table 5. Histologic Findings in Women With a Positive Symptoms Index (n=27)

Histologic Finding	No. of Women
GC/LMP, granulosa cell tumor/tumor with low malignant potential.
Primary ovarian cancer	6
Malignant	4
GC/LMP	2
Simple serous cyst	11
Cystic teratoma	1
Hemorrhagic cyst	1
Fibroma/thecoma/Brenner tumor	1
Leiomyomata	2
Hydrosalpinx/paratubal	2
Other	3

Table 6. Tumor Stage

Group	No. of Patients (%)
	Early Stage (I/II)	Late Stage (III)
TP indicates true positive.
Selection group (n=450)	37 (75.5)	12 (24.5)
Study group (n=272)	23 (76.7)	7 (23.3)
Informative symptoms positive: TP	6 (100)	0 (0)

Because symptoms that persisted for >12 months could be more likely to occur on the first screening encounter than on subsequent annual return screens, results originating from the entry screen were compared with results originating on subsequent screens. There were no significant differences in TP, false‒positive (FP), true‒negative (TN), or false‒negative (FN) results that were identified on the entry screen versus subsequent screens (Table 7). Consequently, no effect on the symptoms reporting was observed because of the possibility of symptoms persisting for >12 months on the first visit, supporting the examination of symptoms results without regard to the number of screening visits.

Table 7. Predictive Ability of the Symptoms Index on Initial or Subsequent Screening Visits (n=272)*

Prediction Status	No. of Patients
	First Encounter	Subsequent Encounters
* There were no statistically significant differences (chi-square statistic, 2.779; P=.669).
True positive	4	2
False positive	12	9
True negative	116	105
False negative	9	15

Identification of Malignancy: Comparisons With the Symptoms Index Alone and With TVS

Does the symptoms index in conjunction with transvaginal sonography improve the identification of malignancy?

The receiver operating characteristic curve (also know as the ROC curve), which measures the tradeoff between the TP rate (TPR) “benefits” and the FP rate (FPR) “costs,” was used to examine the effectiveness of TVS alone and when combined with the symptoms index (Fig. 2). The TPR and the FPR were defined as noted by Suojanen with regard to correctness.28, 29 Screening performance was graphed at and above each MI score with TVS alone in the selection group (n = 450) and in the study group (n = 272) (Fig. 2, open squares). A line of no discrimination describing a random guess or outcome (ie, flipping a coin) is included on Figure 2 as a dashed diagonal line. For the study group, the distance from the line of no discrimination was greatest at MI scores >4 to 5, a relation that was mirrored in the selection group (Fig. 2). Combining the symptoms index with TVS screening moved the ROC space plot dramatically closer to the line of no discrimination, drastically reducing both the area under the ROC curve and the area between the ROC curve and the line of no discrimination (Fig. 2, open circles). By using a logistic regression analysis, the MI score alone (Fig. 2, open squares) significantly (P<.0001) predicted malignancy with an area under the ROC curve (AUC) probability that the MI score alone would rank a randomly chosen positive result higher than a randomly chosen negative result calculated at 0.816. Adding the symptoms results (Fig. 2, open circles) resulted in an insignificant improvement (P = .08; AUC = 0.820) on this model.

Figure 2. Receiver operating characteristic analysis of transvaginal sonography (TVS) and symptoms analysis is shown. The performance of TVS alone is represented by solid circles for the selection group and by open squares for the study group. Open circles represent the discrimination of malignancy by both TVS and symptoms together. The points were plotted with respect to malignancy index values greater than or equal to the number above each symbol. The diagonal dashed line (line of no discrimination) describes a random guess or outcome. Ranges for sensitivity and specificity were obtained from the 2007 report by Goff et al6 and were used for maximal and minimal calculations (the box marked Published Symptoms Data). Solid squares at a true-positive (TP) rate (TPR) between 0.8 and 0.9 near origin of the false-positive (FP) rate (FPR) superimpose the overall results from TVS in the entire University of Kentucky screening group (for sensitivity and specificity source values, see Table 9, Group J). Sensitivity = TP/(TP + FN) = the positive rate = TPR. FPR = 1−specificity. TPR and FPR were defined according to Suojanen.28, 29

The performance of symptoms alone for identifying malignancy (Table 8, Group A) was examined in relation to TVS alone in the study group using the 2 dichotomizations shown in Figure 2 that had the best ROC performance (ie, Group B [MI score ≥4] and Group C [MI score ≥5]) (Table 8). TVS (based on either the Group B or the Group C dichotomization) had much higher sensitivity and lower specificity than the symptoms index. Combining the symptoms index with TVS (Table 8, Groups D and E) in series improved specificity but reduced sensitivity. The results reported here with the symptoms index (Table 8, Group A) resulted in lower sensitivity (Table 8, Group H) and higher specificity than reported by Goff et al6. Because sensitivity and specificity are derived independently of each other, we examined the range of sensitivities and specificities reported by Goff et al6 (Table 8, Group H) in an ROC curve context, Figure 2 (the boxed window identified as “Published Symptoms Data [Goff]”). The published symptoms results underperformed TVS alone, and a portion of these results reached and crossed below the line of no discrimination. The performance of TVS for ovarian cancer screening is summarized in terms of the large group published results (Table 8, Group I) and subsequent accrual results up to June 2008 (Table 8, Group J), both of which indicate better sensitivities and specificities than those produced by the symptoms index alone (Table 8, Group A) or in combination with TVS (Table 8, Groups D and E). Finally, decision making was examined on the basis of meeting either symptoms or TVS criteria in parallel. In this context, TVS results for scores of both MI scores ≥4 and MI scores ≥5 were explored to broadly test the information added by symptoms. Only a minor increase in sensitivity occurred when either TVS MI scores ≥4 or the symptoms index was applied in parallel (Table 8, Group F vs Group B). Similarly, a minor increase in sensitivity occurred when either TVS MI scores ≥5 or the symptoms index was applied in parallel (Table 8, Group G vs Group C) with concomitant decreases in specificity when symptoms were applied with TVS MI scores ≥4 or ≥5. To explain how these findings were made, 1 TP entered Group F (Table 8) because of a positive symptoms index, and that case was not included in Group B (Table 8) because of an MI score <4. Similarly, 14 TNs in Group B (Table 8) were symptom positive and could not be included in the Group F TNs (Table 8). The results of combining symptoms index information with TVS MI scores (Table 8, Groups D and F) indicated that, when both TVS and symptoms criteria are met, there is a reduction of FP results. This reduction, although it improves the positive predictive value (PPV), is at the expense of identifying TPs so that sensitivity suffers (Table 8). When either TVS or symptoms criteria are met (Table 8, Groups F and G), sensitivity is affected positively, but specificity is compromised, with the ability to distinguish malignant from benign (PPV) becoming similar to that of TVS alone (Table 8, Groups B and C).

Table 8. Performance of Either the Symptoms Index or Transvaginal Sonography for Identifying Malignancies When Used Together in the Study Group (n=272)

Group*	TP	TN	FP	FN	No.	Sensitivity, %†	Specificity, %‡	PPV, %	NPV, %	ACC, %
TP indicates true positive; TN, true negative; FP, false positive; FN, false negative; PPV, positive predictive value (TP/[TP+FP]); NPV, negative predictive value (TN/[TN+FN]); ACC, accuracy ([TP+TN]/[TP+TN+FP+FN]); TVS, transvaginal sonography; MI, morphology index. * Screening findings were dichotomized on the basis of an MI≥4 versus an MI≥5. Patients who had an MI≥4 made up Groups B, D, and F; and patients who had TVS findings with an MI≥5 made up Groups C, E, and G. † Sensitivity=TP/(TP+FN)=TP rate. ‡ Specificity=TN/(TN+FP). § The number of patients in the group that received ultrasound. ‖ The PPV lower value was for the entire screening group and the higher value in range was for the period from 2000 to the present, when patients did not undergo surgery for simple cysts.
A: Symptoms positive	6	221	21	24	272	20	91.3	22.2	90.2	83.5
B: TVS, MI≥4	27	135	107	3	272	90	55.8	20.1	97.8	59.6
C: TVS, MI≥5	22	180	62	8	272	73.3	74.4	26.2	95.7	74.3
D: Symptoms positive and MI≥4	5	235	7	25	272	16.7	97.1	41.7	90.4	88.2
E: Symptoms positive and MI≥5	5	237	5	25	272	16.7	97.9	50	90.5	89
F: Symptoms positive or MI≥4	28	121	121	2	272	93.3	50	18.8	98.4	54.8
G: Symptoms positive or MI≥5	23	164	78	7	272	76.7	67.8	22.8	95.9	68.8
Reference basis
H: Goff 20076	—	—	—	—	233§	33-47	61-75	—	—	—
I: TVS: van Nagell 200718	51	24,954	313	9	25,327	85	98.7	14 -27.1‖	99.96	98.7
J: TVS: Up to June 2008	62	31,287	388	11	31,748	84.9	98.8	13.8-20.6‖	99.96	99.8

Finally, when age was overlaid on these considerations, the symptoms index alone appeared to have more sensitivity in premenopausal women (Table 9, Group A) but less specificity than in postmenopausal women (Table 9, Group H). Conversely, TVS alone appeared to have more sensitivity in postmenopausal women (Table 9, Groups I and J) with slightly higher specificity in premenopausal women (Table 9, Groups B and C). When combining symptoms criteria in series with TVS, specificity appeared high in both premenopausal women (Table 9, Groups D and E) and postmenopausal women (Table 9, Groups K and L), but sensitivity appeared quite low. When decision making was examined on the basis of meeting either symptoms or TVS criteria in parallel, sensitivity and specificity appeared to be higher in the premenopausal group (Table 9, Groups F and G vs Groups M and N). However, differences between premenopausal and postmenopausal women were only apparent, because none were significant (P < .05; chi-square test).

Table 9. Age and the Performance of Either the Symptoms Index or Transvaginal Sonography for Identifying Malignancies

Group*	TP	TN	FP	FN	Sensitivity, %	Specificity, %	PPV, %	NPV, %	ACC, %
TP indicates true positive; TN, true negative; FP, false positive; FN, false negative; PPV, positive predictive value (TP/[TP+FP]); NPV, negative predictive value (TN/[TN+FN]); ACC, accuracy ([TP+TN]/[TP+TN+FP+FN]); TVS, transvaginal sonography; MI, morphology index. * Pair-wise chi-square tests were not significant at P < .05 (PAH=.03311; PBI=.3593; PCJ=.5795; PDK=.1053; PEL=.2228; PFM=.6299; PGN=.8399).
Premenopausal (N=45)
A: Symptoms positive	1	36	6	2	33.3	85.7	14.3	94.7	33.3
B: TVS, MI≥4	2	25	17	1	66.7	59.5	10.5	96.2	66.7
C: TVS, MI≥5	2	32	10	1	66.7	76.2	16.7	97	66.7
D: Symptoms positive and MI≥4	0	39	3	3	0	92.9	0	92.9	0
E: Symptoms positive and MI≥5	0	40	2	3	0	95.2	0	93	0
F: Symptoms positive or MI≥4	3	22	20	0	100	52.4	13	100	100
G: Symptoms positive or MI≥5	3	28	14	0	100	66.7	17.6	100	100
Postmenopausal (N=227)
H: Symptoms positive	5	185	15	22	18.5	92.5	25	89.4	18.5
I: TVS, MI≥4	25	110	90	2	92.6	55	21.7	98.2	92.6
J: TVS, MI≥5	20	148	52	7	74.1	74	27.8	95.5	74.1
K: Symptoms positive and MI≥4	5	196	4	22	18.5	98	55.6	89.9	18.5
L: Symptoms positive and MI≥5	5	197	3	22	18.5	98.5	62.5	90	18.5
M: Symptoms positive or MI≥4	25	99	101	2	92.6	49.5	19.8	98	92.6
N: Symptoms positive or MI≥5	20	136	64	7	74.1	68	23.8	95.1	74.1

DISCUSSION

Our experience from nearly 20 years of screening has positioned us to reach certain conclusions regarding the benefits and limitations of TVS as a screening method for ovarian cancer, especially in relation to safety, patient acceptability, time-efficiency, costs, and interobserver variation.30-32 In this regard, we believed that it would be beneficial to determine how symptoms information could improve TVS screening for ovarian cancer. It is noteworthy that, as the basis for this study, we were in a position to draw on a group of patients all of whom had a sufficient TVS abnormality that led to surgery and histologic evaluation of the ovarian tumor removed.

For a screening test to be effective, sensitivity, specificity, PPV, and negative predictive value should be high.33 With periodic use, screening should decrease stage at detection and should increase disease-specific survival in the screened population.34 TVS may be questioned as a screening method for ovarian cancer because of its moderate sensitivity and relatively low PPV.18 Although TVS screening accurately predicted the presence of cancer involving the ovary in 62 asymptomatic women, there were 11 patients who developed ovarian cancer within 12 months of a normal scan (Table 8, Group J, FN column), so that the addition of information from a symptoms index would have the potential of reducing both the FN and the FP screening results. Such a reduction would have its greatest impact by increasing the PPV so that screening would better differentiate malignant from benign lesions. The results reported herein when both TVS and symptoms criteria are met, indicate a reduction of FP results. This reduction, although it improves the PPV, is at the expense of identifying TP results, so that sensitivity suffers (Table 8). When either TVS or symptoms criteria are met, sensitivity is affected positively, but specificity is compromised, and the ability to distinguish malignant from benign (PPV) becomes similar to that of TVS alone (Table 8).

The clinical significance of the results reported here is that a screen that is negative by both ultrasound and the symptoms index is highly likely to be benign (specificity, >97%). Second, adding in parallel symptoms information with equal weight to that of ultrasound (either/or) only slightly improved the discrimination of malignancy (1 additional TP with a sensitivity increase of +3.3%). These results strongly indicate that the major screening benefit in discriminating malignancy is achieved with ultrasound findings, whereas symptoms information can aid in reducing surgery for women who have benign conditions that generate an ultrasound abnormality.

A distinction between the data reported here and those reported by Goff et al6 is that they differ in composition: the Goff et al study was composed of 37% early stage disease versus 75% in the current study. This variation in stage distribution alone could account for differences between the reports in sensitivity and specificity, particularly if symptoms are more apparent in late-stage ovarian cancer. We have used exclusion based on reported patient uncertainty to constrain recall bias, whereas Goff et al6 used symptoms queries limited to the present. Even more significant is the overall difference in the selection of the study group. The study group that we used in the current report was composed of women from the general population who were actively participating in the University of Kentucky Ovarian Screening Program. In contrast, Goff et al6 constructed a study group that surveyed 149 women with ovarian cancer (55 early stage, 88 late-stage, 6 unknown stage), 225 high-risk women who were enrolled in an ovarian cancer early detection study, and 233 women with conditions indicating that they should receive an ultrasound. It is entirely possible that this construction does not mirror the same cross-section of women with and without symptoms that we observed in our large screening effort in the general population. In addition, Daly and Ozols35 have presented a commentary on issues that may involve limitations of the control group used by Goff et al.6 An additional consideration is presented in the 2008 report by Anderson et al that combines the Goff symptoms index with CA 125 to improve detection of ovarian cancer.36 In that report,36 sensitivity and specificity of the symptoms index alone were considerably higher than first reported.6 The ranges reported initially were 33% to 47% sensitivity and 61% to 75% specificity6; whereas, in the recent report,36 the ranges were 52.1% to 74.8% for sensitivity and 83.6% to 91.9% for specificity. These differences raise several questions. First, because the sample for the recent study is described as a subgroup of the previous study, indicating that different study groups were not used, what accounts for the increased sensitivity and specificity in the recent report? Second, if the subgroup that was reported36 differed intrinsically from the originating group from which it was drawn,6 does this indicate that there will be considerable variability in the results obtained with the symptoms index? Although there are differences in study design, our findings indicate that the evaluation of informative symptoms as identified by Goff and collaborators6 can be useful in helping distinguish benign disease when used in conjunction with TVS. Symptoms information cannot be ignored and is an important element in communication between patient and physician. The efforts of Goff et al6 have done much to erode the idea that ovarian cancer has no symptoms until it is advanced. In this regard, efforts still are needed to define how this information frequents and affects the general population.

With regard to the clinical outlook, the following considerations are evident: 1) TVS alone performs better than informative symptoms evaluation in identifying malignancy, whereas informative symptoms evaluation in concert with TVS improves sensitivity only modestly. 2) The frequency of informative symptoms cannot be dismissed because, even when they are present in only 20% of ovarian malignancies, the risk for these patients is higher than that for patients without informative symptoms. 3) The evaluation of informative symptoms is facile, in that it is simple, quick, and inexpensive.

Taken together, these considerations point in a clear clinical direction. TVS performs with credibility in women who are both symptom-free and in those who report symptoms. By itself, TVS performs as well as or better than mammography,37 mammography-assisted by sonography,38 or mammography combined with magnetic resonance imaging,39 even with regard to PPV. In this sense, TVS lives up to the performance standards used and accepted for screening. In a population in which informative symptoms accompany malignancy in 20% of cases, informative symptoms evaluation in parallel equally weighted with TVS (Table 8, Groups F and G) carries a 3% to 4% improvement in detection (sensitivity). However, for the additional positive case that is detected, the data reported here indicate that from 4 to 14 additional women without malignancy will undergo surgery. The information that needs to be emphasized is that we observed 6 malignancies in 27 women who were positive for informative symptoms (22%) and 24 women with malignancies that were absent informative symptoms among 30 women who had malignancies (80%). It is possible that the informative symptoms coincided only serendipitously with malignancy; however, this can be neither substantiated nor refuted by the current study. Moreover, in the study reported here, all women had a successful TVS. Performing a definitive TVS should be the first line of response when encountering a patient who has informative symptoms. Unfortunately, TVS may not be reliable in clinical settings if a large uterus is present or if the patient's BMI is excessive. In these cases, thorough assessment of a patient with informative symptoms using other diagnostic testing is essential. It is important to educate patients that informative symptoms should not be ignored and that the degree to which symptoms are a resultant indicator of early stage ovarian malignancy has yet to be determined.

In the absence of early detection, most patients with ovarian cancer will continue to present with advanced-stage disease in which the cost of treatment is high and the cure rate is low.40 We intend to continue to use symptoms analysis prospectively in combination with TVS to evaluate more fully the performance of screening for ovarian cancer. As specific biomarkers are added to TVS ovarian screening protocols that include symptoms analysis, the protection afforded by annual screening from ovarian cancer mortality should improve.

Conflict of Interest Disclosures

Supported by grants from the Telford Foundation, and the Department of Health and Human Services, Commonwealth of Kentucky.