"Benign" Tumors and "Early Detection" in Mammography-Screened Patients of a Natural Cohort With Breast Cancer FREE

BREAST CANCER screening, which began to include the widespread use of mammography (MMG) in the early to mid-1980s, has been studied many times since the early 1960s, including 8 randomized controlled trials and subsequent reviews and meta-analyses.^1- 27 The results of the studies have been mixed, but they generally suggest that screening leads to decreased mortality from breast cancer. The results have raised 2 important problems in interpretation. The first is that MMG often discovers nonpalpable tumors, having diverse histological types and degrees of biological aggressiveness,^28- 30 the impact of which is not clearly understood. The second problem occurs because tumors are almost always staged according to the size and extent of disease. Consequently, the tiny tumors (<2 cm) detected by MMG screening will usually be assigned a lower TNM stage, and the preponderance of tumors in good prognostic stages may then artifactually raise the survival rates associated with MMG.

The performance of MMG could be better evaluated if the biological behavior of tumors were better classified, so that slow-growing or indolent cancers, which might not kill patients if left alone, were analyzed separately from the virulent tumors whose removal might be lifesaving. This biological distinction cannot be discerned with the customary staging categories based on gross or microscopic morphological characteristics, but might be aided if the cancers were also classified according to their mode of detection, as having been discovered with or without screening.

This study was done to evaluate the latter possibility and to reexamine the roles of early detection, cure, and other factors associated with screening in improving survival for breast cancer.

A cohort of women with breast cancer was assembled through the Tumor Registry files of Yale–New Haven Hospital (YNHH), which is both a community hospital and a referral center in New Haven, Conn. To be eligible for inclusion, each woman must have had a histologically demonstrated, primary carcinoma of the breast, diagnosed while she was alive, that received its first antineoplastic treatment at YNHH between January 1 and December 31, 1988. This timing had 2 advantages: it allowed at least a 5-year follow-up to be determined for all patients when this investigation began in 1994; and 1988 was one of the first years in which MMG was both widely used for screening and also had high technological quality in detecting very small tumors. These criteria for eligibility would produce a cohort that was "natural," since it contained all women whose breast cancer was first treated at YNHH during the cited year. No selection criteria were used to produce special subgroups defined by demographic, morphological, or other distinctions. At YNHH, MMG was performed with a standard film-screen technique on an MMG unit (Senographe 500T; Thompson CGR Medical, Columbia, Md), with the use of Kodak Ortho M film (Eastman Kodak Co, Rochester, NY) and MIN-R screens (Eastman Kodak Co).

The clinical status and course of each patient were ascertained primarily from the original medical records and secondarily from the YNHH Tumor Registry files. When the information in these archival sources was insufficient, letters were sent to private physicians and, if necessary, the patients were interviewed by telephone. All procedures were reviewed and approved by the Human Investigations Committee at Yale University School of Medicine.

Zero time for each patient was defined as the date of the first antineoplastic therapy, which consisted of surgery, radiotherapy, or chemotherapy. For zero time status, patients were classified according to age, race, insurance type, family history, presence of symptoms, comorbidity, stage of reproductive cycle, estrogen and progestogen status, mode of detection (with or without MMG screening), tumor histological features, nodal involvement, anatomic TNM stage, initial treatment, and subsequent clinical course. The data were coded, by specific criteria, to prepare for computerized analysis.

Diagnosis by MMG screening was defined as a "positive" mammogram obtained before the patient or the physician was aware of an abnormality in the breast or axilla. For this designation, the mammogram had to have been marked as a screening test; the patient must not have had any complaints of breast symptoms, such as breast pain or tenderness, abnormal nipple or skin changes, or a lump discovered on breast self-examination; and the physician must not have noted a suspicious abnormality in the breast or axilla during any preceding clinical examination. The MMG screening was documented if the test was done because the woman received an annual reminder in the mail, or attended the Yale Mammography Van either when it was in the neighborhood or by referral from her physician for a routine test.

A cancer not detected by MMG screening could have been found via other screening, if the physician or an otherwise asymptomatic patient discovered a lump on a breast examination that was done routinely, without any previous evidence that the lump was present.

The mode of detection was classified as symptomatic if the lump was previously known to be present, having been noted at an earlier examination, or if the cancer was found when a woman came to her physician with complaints of one or more of the breast symptoms cited earlier.

Because the patients' histories were not always adequately recorded, we classified only the immediate mode of detection for the cancer at zero time. We could not accurately determine whether the screenings represented a first episode or an "interval" detection after an earlier negative examination.

The anatomic TNM stages at zero time were classified according to the 1988 American Joint Committee on Cancer Staging scheme for breast cancer.³¹ The details of the complex arrangement used for staging are depicted in Figure 1, which shows combinations formed from the size of the primary tumor vertically in the rows and from the degree of metastases in the columns. The tumor sizes are listed as tiny or small (≤2 cm), moderate (>2 to ≤5 cm), and large (>5 cm). The metastases are divided mainly according to ipsilateral axillary node(s), regional, and distant involvement. The ipsilateral nodes are further divided as movable or fixed, and the movable nodes have either micrometastases or macrometastases. Regional involvement includes contralateral lymph node(s), ipsilateral internal mammary lymph node(s), or direct extension to the chest wall or skin. Distant involvement includes ipsilateral supraclavicular lymph node(s) as well as extrathoracic spread.

In stage I, the primary tumor is tiny or small, with no evidence of metastasis. In stage IIA, a small (or absent) primary tumor has either micrometastasis (<2 mm) or macrometastasis (≥2 mm) to movable ipsilateral axillary node(s), or a moderate-sized tumor has no metastasis. Stage IIB consists of a moderate-sized tumor with metastasis (micrometastases or macrometastases) to movable ipsilateral axillary lymph node(s), or a large tumor without metastasis. Stage IIIA contains primary tumor of any size with metastasis to fixed ipsilateral axillary node(s), or a large tumor with metastasis to movable ipsilateral node(s). Stage IIIB consists of any size primary tumor with regional involvement, and stage IV contains any size primary tumor that has distant metastasis.

Carcinoma in situ (CIS) consists of microscopic lesions confined to the ducts and lobules of the breasts.^29- 30 Pathologists regularly divide CIS into ductal and lobular categories; the lobular form is sometimes regarded not as a precursor lesion, but as a risk factor for cancer.³² Cancers are called in situ if confined to the basement membrane but are considered microscopically invasive if the integrity of the basement membrane is breached. Although these presumed cancers are discovered mainly with MMG screening, some are large enough to be palpated on physical examination. All of the in situ cancers are classified as stage 0 (CIS) regardless of tumor size.

Outcome events were classified in the following categories: death from breast cancer, from comorbid disease, or from an unknown cause; or alive with or without recurrence of the initial breast cancer. For life-table analysis, the failure event was a cancer recurrence or a cancer death. In the absence of a cancer recurrence, patients who died of a comorbid disease or who were unavailable for follow-up were classified (in the conventional manner) as "withdrawn alive." Cancer recurrence was defined as the discovery of a tumor of the same previous histological type after the patient had been deemed free of cancer after completion of treatment. A newly discovered tumor was considered a second primary, rather than a recurrence, if it had a different histological type or occurred in the contralateral breast.

The cause of death was determined mainly from recorded clinical information about events leading to the patient's death. When the medical records were inadequate, the cause of death was ascertained from the patients' private physicians, from death certificates obtained from local state health departments, or from a national death registry. Deaths were attributed to breast cancer if breast cancer had been stated as the cause of death in the death certificate, medical record, or physician's record. (In almost all instances, these deaths were preceded by evidence of metastatic breast cancer.) Otherwise, death was attributed to comorbid disease. Decisions about cause of death were almost always made without awareness of mode of detection, staging, or antecedent therapy.

Tumor-free survival was calculated as the duration from the date of zero time to the date of tumor recurrence or death from breast cancer.

After the coded data were verified, the statistical analysis was done with SAS stastical software (Version 6.12; SAS Institute Inc, Cary, NC). Baseline demographic and clinical characteristics underwent bivariate analysis. Survival and recurrence rates were calculated by the life-table method, with comorbid deaths analyzed by regarding the patient as withdrawn alive at the time of death. The statistical difference between Kaplan-Meier curves was analyzed with the log-rank test. Multivariable adjustment of survival curves was performed with proportional hazards (Cox) regression analysis.

Of the 333 medical records initially reviewed, only 257 were eligible for the current research. (In 75 patients, the first antineoplastic treatment occurred in a year other than 1988 or was given at an institution other than YNHH, and 1 patient did not have histologically proved cancer.) Of the 257 eligible patients, 24 (9%) were excluded for the following reasons: 2 had no follow-up beyond zero time; medical records could not be obtained for 12; and in 10, the data regarding mode of detection or staging were inadequate. For the remaining 233 patients, who constituted the study cohort, the median length of follow-up was 82.4 months; only 8 did not have follow-up data for at least 5 years.

Table 1 shows that the cohort had a median age of 62 years, with a range of 26 to 87 years; most were white (208 [90%]) and had private insurance and/or Medicare (191 [93%]). In clinical characteristics, 14% of patients had first-degree relatives with a history of breast cancer; 78% were in TNM stages 0, I, or IIA; and only 8% had a Charlson Comorbidity Score³³ greater than 2. The patients were mainly postmenopausal (68%), and 78% had ductal histological type, with positive hormonal receptors (when measured) in 67% (133/200) for estrogen and 59% (104/177) for progestogen.

The method of breast cancer detection was by MMG screening in 42%, other screening in 40%, and symptomatic manifestation in 18%. Three patients who had incidental discoveries of their breast cancer were listed in the other-screening group, since their cases were neither appropriately symptomatic nor detected by screening MMG. One patient had a spider bite and developed a rash under her breast; examination for the rash disclosed a 0.5-cm lump. A second woman, who underwent mammoplasty for breast reduction, was found to have CIS in the examined tissue. In a third, who was being followed up for benign-appearing microcalcifications on an earlier screening mammogram, a satellite CIS was discovered incidentally.

The previously cited demographic and comorbid attributes had essentially similar distributions among the 3 detection groups, although the other-screening group was younger, with a median age of 55 years, compared with 64 years in each of the other 2 groups. The other-screening group also had fewer patients known to be postmenopausal (55% [51/92]) compared with the symptomatic groups (77% [34/43]).

Figure 2 shows survival curves for the proportion of patients free of recurrence or cancer death, classified in 5 TNM groups for which the small numbers in stages III and IV have been combined. The 5 curves have obvious quantitative differences that are statistically significant (log-rank χ²=91.2; 2P<.001), and the trends are consistent throughout, without any crossovers of curves. The results remained essentially unchanged when adjusted for age, menopausal status, white race, and/or insurance status. Patients in TNM stage 0 (CIS) had no recurrences or cancer deaths, and the curves were similar for TNM stages I and IIA, which allowed those 2 stages to be combined for further analyses.

As shown in Table 2, the proportion of patients who had their breast cancers detected by MMG screening was 77% (24/31) in the CIS group; 53% (50/94) and 28% (16/57), respectively, in the good-prognosis TNM stages I and IIA; 8% (3/36) in stage IIB; and 27% (4/15) in the combined TNM stages III and IV.

Table 3 shows the outcome events in the 5 TNM groups, divided into the 3 modes of detection for the cancers. The "failure events" for life-table analyses consisted of 4 possible events, which were counted only once: cancer death and no previous recurrence, cancer death and recurrence, comorbid death and recurrence, and still alive with recurrence. The withdrawals group contains patients without recurrence who had a comorbid death, were unavailable for follow-up, or were still alive at the end of the observation period.

Of the 31 patients with CIS, none had recurrences or cancer deaths, regardless of how the breast cancers were discovered. (Six of the 31 cases were lobular and 25 were ductal.) In stages I and IIA, none of the women with cancers detected by MMG screening had cancer deaths, and only 1 had a recurrence, but 11 of the women detected with other screening or symptoms either died of breast cancer or had a recurrence. In stage IIB and in the combined stages III and IV, the MMG screening–detected group also had better outcomes than all others.

Table 4 demonstrates the downward shift in TNM stage expected with screening. The tumors discovered by MMG screening are predominantly (90 [93%]) in the lower TNM stages 0, I, and IIA. In the other detection groups (other screening or symptoms), the cancers were shifted toward the higher stages. This shift was not unexpected because the larger tumors are often palpable. Among the other-screening group, about 60% of tumors were found by palpation in stages IIA and above. A similar pattern was seen for the symptomatic women, but the numbers were too small for firm conclusions.

The clinical impact of the foregoing results is further demonstrated in Figure 3, which shows survival curves for the combined TNM stages I and IIA, stratified by method of detection. The other-screening and symptomatic detection groups had similar results, but the MMG screening group had distinctly higher survivals (log-rank χ²=6.50; 2P=.04). These results were also essentially unchanged with multivariable adjustments for age, menopausal status, race, and/or insurance status. The findings demonstrate that MMG screening seems to detect tumors that are more benign than those found by other methods. Furthermore, other screening provided no improvement in outcomes when compared with the symptomatic group. (Because the other-screening and symptomatic groups had similar survival curves, these groups were combined for subsequent analyses of therapeutic decisions.)

Table 5 shows the type of antineoplastic therapy received by women in stages I and IIA, stratified by the method of cancer detection. The 3 main types of surgical treatment are marked as "only" if used alone and as "plus" if accompanied by radiation therapy and/or chemotherapy. Although all patients had surgery, the rates of the 3 progressively more aggressive forms of surgical therapy were, respectively, 26%, 27%, and 47% in the MMG screened group, compared with 15%, 25%, and 60% in all others. The total rates of augmented ("plus") therapy were 42% in the MMG screened group and 41% in all others. Since the MMG screened group was not treated more aggressively, treatment cannot be held responsible for the group's better survival.

For the 31 women with CIS, who all had event-free survival, Table 6 shows that aggressive therapy, ie, more than lumpectomy alone, was used in 23 (74%) of the patients, and that 15 (48%) had mastectomies, including 3 who underwent bilateral partial mastectomies. Since none of these patients had cancer death or recurrence, regardless of the extensiveness of treatment, the need for aggressive forms of therapy might be reconsidered.

These results should not be used to question the value of MMG screening. Its comparative efficacy is best discerned from randomized trials, particularly those in which the main outcome is a reduction in all-cause mortality.

The main findings of this study of a natural cohort of unselected patients offer confirmatory evidence for 2 suspicions about the spectrum of biological behavior in breast cancer. First, the spectrum has rapidly growing lethal components as well as others that grow slowly in a relatively benign manner.³⁴ Second, the slow-growing lesions are particularly likely to be found in groups detected by MMG screening. (The term length bias is sometimes applied³⁵ to the idea that slow-growing cancers are particularly likely to be detected by screening, but the term does not include the possibly separate category of cancers so "benign" that they may not grow.)

The mammographic production of early detected cases is shown particularly well in Table 4, where the stages with a good prognosis (CIS, I, and IIA) were found in 93% of the cancers detected by MMG screening, compared with 73% in those found by other screening and with 53% in the symptomatic group. The detection of relatively benign tumors is shown well in Table 3 and in Figure 3, where patients within the same TNM stage had better outcomes if detected by MMG screening than by the other modes of discovery.

The relative benignity of the MMG-screening–detected tumors is also demonstrated by the predominance of CIS lesions in this group. The biological behavior of these lesions is controversial, and some writers³⁶ have suggested that they should not be regarded as cancer. In our own series, no recurrence or any cancer problem developed during the follow-up period in any of the 31 patients with CIS, regardless of mode of discovery. Although a longer follow-up period in a larger case series would be needed to confirm these results, they already provide strong support for women (and physicians) who prefer treatment with simple lumpectomy, rather than more extensive surgery and/or aggressive additional therapy.

Even without the CIS group, however, MMG screening seemed to detect, in each TNM stage, tumors that were more biologically benign than those found with other modes of detection. The concept of a relatively benign form of breast cancer is not new. Several autopsy and other studies^37- 40 have found a substantial reservoir of occult breast tumors that do not spread or kill. In autopsy studies, the prevalence of occult CIS ranged from 4.3% to 18%,³⁸ and it was 1.8% for "invasive" breast cancer.⁴⁰ Doll and Peto⁴¹ contend that these cancers are "histologically malignant, but biologically benign,"⁴¹ although other investigators^32,42 claim that invasion or spread may occur in patients followed up for periods of at least 15 years.

When the "natural duration" of breast cancer after refusal or denial of surgical treatment was examined many years ago,^43- 50 many women survived much longer than expected, and their mean duration of life with untreated breast cancers ranged from 30.2 to 40.5 months⁴⁷ from the onset of symptoms to death. One study⁴⁵ showed that "in those [untreated] patients in whom cancer was detected early and in whom operation was advised, the average duration of life was 42.8 months [compared with 40.5 for the entire cohort]." All stages (I, II, and III) were represented in the group⁴⁵ who refused therapy; both fungating and ulcerating cancers were noted; and several women survived for as long as 5 decades. All of these relatively benign cancers of the past are still available for early detection in the current era of advanced mammographic and other technology.

The current study has the obvious limitation of small numbers. Obtaining the necessary information about mode of discovery for each breast cancer is a time-intensive procedure that requires careful review of each patient's complete medical record, often supplemented by additional inquiries to physicians and patients. Nevertheless, we believe that a relatively small study supplying the desired information could complement the customary larger studies in which the requisite information is absent. Furthermore, although not obtained for all the results, statistical significance was found for the main findings here, and all of the quantitative trends went in consistent directions.

Although the value of MMG screening is not questioned, the results indicate the need for physicians and patients to be wary of overenthusiastic interpretations of therapeutic "success," an error that has occurred in every era of medicine.^51- 52 Since MMG screening detects early-stage and relatively benign tumors, its survival advantages may become substantially overestimated.

Our main conclusions and recommendations are for more extensive research to confirm biological behavior both for in situ lesions and for the more distinctive cancers detected with MMG screening. If many of these cancers are indeed relatively benign, they can be treated less aggressively than in the past.

Accepted for publication June 29, 1999.

This work was done when Dr Moody-Ayers was a Robert Wood Johnson Clinical Scholar at Yale University School of Medicine, New Haven, Conn.

We thank Susie Roberts for her assistance and efforts in procuring the subjects' medical records, Maria Johnson for preparing and sending out letters of inquiry, and Sharon Inouye, MD, for suggestions regarding data presentation.

Reprints: Alvan R. Feinstein, MD, MS, Yale University School of Medicine, Room I-456 SHM, PO Box 208025, New Haven, CT 06520 (e-mail: arf2@e-mail.med.yale.edu).