Identifying Breast Cancer Recurrence Histories via Patient-Reported Outcomes

Abstract

Purpose:

To test accuracy of patient self-report of breast cancer recurrence for enhancing standard population-based cancer registries that do not routinely collect cancer recurrence data despite the importance of this outcome.

Methods:

Potential research subjects were identified in the Breast Cancer Research Database (BCRD) of the Swedish Cancer Institute (SCI). The BCRD has collected data within 45 days of each medical encounter on new primary breast cancer patients receiving all or part of their initial care at SCI. Females diagnosed with a new primary breast cancer 2004–2016, Stages I-III and alive at the time of study initiation (2018) were identified. Recurrent breast cancer patients were matched 1:1 to surviving non-recurrent patients by patient age, date of diagnosis, and single or multiple primary tumors. Consented research subjects were surveyed about their initial and subsequent diagnostic, therapeutic and recurrent events. PRO survey responses were compared with BCRD information for each individual participant. Discrepancies were reviewed in medical records.

Results:

A matched sample of 88 recurrent and 88 non-recurrent patients were used in analyses. Respondents correctly identified the date of diagnosis of first primary breast cancer within 1 year 94% (165/176). Recurrence was reported by 97% (85/88) of recurrent patients. No recurrence was reported by 100% (88/88) of non-recurrent patients. Recurrence date within 1 year was correctly identified in 79% (67/85). Recurrence site was correctly identified in 82% (70/85). Medical record review of survey-registry discrepancies led to BCRD corrections in 4.5% (8/176) of cases.

Conclusions and Implications for Cancer Survivors:

Breast cancer patients can accurately report their disease characteristics, treatments and recurrence history. Patient-reported information would enhance cancer registry data.

Introduction

Cancer registries are indispensable for tracking the evolving burden of cancer in the population. These registries maintain information on cancer incidence, patient demographics, disease characteristics at diagnosis, initial treatments, and mortality for newly diagnosed patients. In the US, the Surveillance, Epidemiology, and End Results (SEER) Program (https://seer.cancer.gov/) of the National Cancer Institute (NCI) facilitates data collection at diagnosis in 18 geographically defined catchment areas. SEER consolidates the information for population studies. Limited information on cancer recurrence or progression is routinely collected.

Cancer recurrence has a major influence on patient survival. It is a key event triggering further treatment decisions and represents a significant worry for patients. Lack of reliable recurrence information has led to explorations of other approaches to determine the risk and frequency of cancer recurrence. For aggregate summaries, Mariotto¹ decomposed disease-specific survival from diagnosis for non-distant metastatic cases into the time from diagnosis to metastasis and the time from metastasis to death. By using external estimates of survival from metastasis to death, they were able to estimate the distribution of the time to disease recurrence in populations of stage I, II and III patients. For individual-level data, statistical learning and data mining approaches have been harnessed to predict recurrence events from claims histories. Chubak² used classification and regression tree (CART) analysis to predict whether and when a patient had experienced a breast cancer recurrence or second breast cancer diagnosis. Ritzwoller³ used a combination of logistic regression and changepoint detection to identify the presence and timing of recurrence events. Recently, A’mar and coauthors⁴ applied a similar changepoint idea, but to the single series of monthly predicted probabilities of being post recurrence. This accommodated a large number of novel features that leverage the month-based definition of the statistical learning problem.

Recently, Patient Reported Outcome (PRO) measures have been developed targeting specific treatments and interventions.^5,6 PRO studies have explored subjective information (e.g. quality of life, symptoms, psychological effects, physical findings, sexual wellbeing, complications, functional activity, social relationships, spirituality and satisfaction).^7–22 PRO measurements have enhanced health care decision-making throughout the cancer control spectrum from prevention and diagnosis through early and long-term treatment and management to routine survivor, hospice and terminal care programs.

This study examines the potential of patient self-report to enhance the objective record of cancer recurrence for use in cancer registries and research. We use a long-standing Breast Cancer Research Database (BCRD) which prospectively records information regarding breast cancer diagnosis, disease characteristics, initial treatments, breast cancer recurrence and death. We identified breast cancer survivors with and without cancer recurrence histories and surveyed them about their breast cancer diagnosis, characteristics, initial treatments, and recurrence history. We compared survey results with the ‘gold-standard’ BCRD information.

Methods

The BCRD was established in 1989 at the Swedish Cancer Institute (SCI). The SCI is a multidisciplinary oncology specialty unit of the Swedish Medical Center (SMC) in Washington State. BCRD data are collected on each newly diagnosed breast cancer case that has all or part of the initial treatment of the breast cancer at the SCI. The BCRD cases represent a subgroup of the SMC’s Cancer Registry analytic cases with optimal access to medical information at the SCI. In addition to standard cancer registry information, the BCRD also prospectively collects data regarding recurrence of the breast cancer and subsequent treatments. The BCRD is managed full-time by an experienced cancer registrar. The data are collected from the institution’s medical record (currently electronic) within 45 days of an encounter and each case is monitored for future patient encounters. Information regarding care at other institutions is obtained directly from providers’ offices, other institution medical records and the Cancer Surveillance System (CSS), the SEER regional catchment area database maintained at the Fred Hutchinson Cancer Research Center (FHCRC). The BCRD maintains follow up of over 95% of cases. In June 2018 when this study began, the BCRD contained 17,781 cases.

Female cases diagnosed with AJCC stage I-III invasive breast cancer between 2004 and 2016 inclusive (n=8,222) were identified. For patients with multiple primary breast cancer diagnoses, the index breast cancer case was the last primary breast cancer diagnosed. Patients with a new non-breast cancer after the index breast cancer diagnosis, patients with prior stage IV cancer and deceased patients were excluded. Of the remaining 6430 breast cancer diagnoses, 5487 had only one and 947 had multiple breast cancer diagnoses.

We used a matched design to select survey participants. We identified all surviving patients who had had a recurrence after their index breast cancer and the medical record of these potential survey subjects was reviewed for eligibility. Those undergoing active care for advanced disease were reviewed with the treating oncologist. This case screening was performed to confirm subject eligibility and to identify potential subjects whose ongoing medical care or well-being would be compromised by survey participation.

For each recurrent subject, non-recurrent potential subjects were identified with index breast cancer diagnosed within one year, age at diagnosis within 5 years and the same single or multiple cancers. The non-recurrent potential research subjects matching to each recurrent research subject were listed in a random order. Screening of the non-recurrent potential research subjects consisted of medical record review only. Screening and contact of non-recurrent potential research subjects started upon enrollment of the matched recurrent subject.

The Patient-Reported Outcome Measurement Information System (PROMIS®) standards, guidelines and examples (https://www.nia.nih.gov/research/resource/patient-reported-outcomes-measurement-information-system-promis) were used in drafting the survey. Advice was provided by 3 experts in the generation and utilization of patient-reported surveys. The draft surveys were reviewed and tested by a breast cancer patient/survivor focus group and further modified. The final survey was brief, required a short time to complete (2–5 minutes) and did not engender significant anxiety in focus group participants or survey experts.

All study procedures were approved by the SMC Institutional Review Board (IRB). The survey, the Informed Consent Form, and Talking Points and Verbal Presentation materials for clinical and research staff education and training were each approved by the IRB and are included in the Appendices. [dummy]

The survey had 7 questions pertaining to general health, initial breast and subsequent cancer diagnoses, treatments, and recurrence. The survey questions explicitly distinguished new breast cancers and breast cancer recurrence following diagnosis. Our objective was to determine whether survey respondents could provide accurate information about their recurrence histories and distinguish recurrence of breast cancer from a new breast or other cancer. Once questions about diagnosis history had been asked, the index cancer diagnosis and its diagnosis date were provided to the patient for reference in the rest of the survey.

SCI breast cancer clinicians were provided with the study information and potential research subjects were monitored by study personnel for scheduled clinic visits. Potential research subject contact was in-person or by phone. When possible, a research staff member met with the patient in the outpatient area after introduction by the provider or clinical team, outlined the study, reviewed the informed consent form (ICF) and answered questions. Patients who volunteered to participate in the study and signed the ICF were given a copy of their signed ICF. Enrolled subjects were offered to complete the survey in the clinic on paper, on a tablet in REDCap, or by providing the survey answers verbally to the research staff member for transcribing on a paper survey copy. All paper surveys were promptly entered into the REDCap database.

Potential subjects no longer having appointments at SCI within 2 months and those moved to other locations were contacted by phone and given a verbal outline of the study with review of the informed consent material. Upon request, study documentation was sent to potential participants for more in-depth review. The phone answers were transcribed by the research staff member on a paper survey. Voicemail messages with a call back phone number were usually left for unanswered calls to facilitate future phone contact. Those who did not return the call were phoned again in 1 week to 1 month. At least 3 calls with voicemail messages were left for potential recurrent subjects and 1 for potential non-recurrent before the research subject was declared as ‘no response’.

Where survey answers diverged from the BCRD data, the BCRD and medical record were examined, and the CSS data were often reviewed to clarify discrepancies. When the patient-reported answer appeared to be correct, the case was reviewed with the manager of the BCRD. The BCRD was updated or corrected as appropriate and was then used in all analyses. Survey answers were not modified.

Fisher’s exact test was used to test nonrandom association between two categorical variables. It does not provide standard error or range. Sensitivity, specificity, positive predictive value, negative predictive value and accuracy of recurrence self-report was calculated.

Results

We identified 189 (150 single and 39 multiple breast cancer) patients with recurrence potentially eligible for survey participation. Screening removed 65 cases: 6 lost contact, 34 deceased, 8 incapacitated, 14 informed consent in English only, 2 oncologist recommends no contact, and one case with multiple breast primaries listed twice (BCRD error); leaving 124 eligible patients. Of these, 34 of 38 (89%) met in clinic enrolled and completed the survey (4, 11% declined); and 54 of 68 (79%) contacted by phone completed the survey, (14, 21% declined). No contact or response to voicemails occurred in 21% (18/86) of patients called. Of recurrent cases contacted 83% (88/106) participated in the survey (18/106, 17% declined).

Of 26 non-recurrent subjects contacted in clinic, 22 (85%) completed the survey (4, 15% declined). Of 86 non-recurrent potential research subjects contacted by phone, 74 (86%) were enrolled (8, 9% declined) and 4 were ineligible (3 incapacitated, 1 BCRD error). Only one voicemail was left for a non-recurrent potential research subject and contact with the next matched subject was initiated after 1–4 weeks. This resulted in duplicate completed surveys in 8 cases. Only the first completed survey was used. No contact or response to the voicemail occurred in 63% (145/231) patients called. Of eligible non-recurrent cases contacted 89% (96/108) participated in the survey (12/108, 11% declined). Overall, enrollment rate for both eligible recurrent and non-recurrent cases was 88% (56/64) for clinic contact and 85% (128/150) for phone contact.

Table 1 summarizes patient, tumor and primary treatment characteristics for the 2 groups. By design, patient age and year of diagnosis and number of cancers (single or multiple) were similar between groups. Recurrent patients had higher stage (p<0.001) and Bloom-Richardson grade (p=0.007), and more frequent total mastectomy (p<0.001) and chemotherapy administration (p=0.006).

Table 1:

Patient, risk factor and primary treatment characteristics within recurrent and non-recurrent groups

Characteristics	Recurrent Group N=88		Non-Recurrent Group N=88		Fisher’s Exact Test *
	N	%	N	%
Age at Index Diagnosis < 50 years	45	51.1%	47	53.4%	p = 0.920
Age at Index Diagnosis 50 – 70 years	35	39.8%	35	39.8%
Age at Index Diagnosis > 70 years	8	9.1%	6	6.8%
Index Diagnosis Year 2004 – 2008	32	36.4%	30	34.1%	p = 0.963
Index Diagnosis Year 2009 – 2012	32	36.4%	33	37.5%
Index Diagnosis Year 2013 – 2016	24	27.3%	25	28.4%
Single Primary Breast Cancer	74	84.1%	79	89.8%	p =0.371
Multiple Primary Breast Cancers	14	15.9%	9	10.2%
Stage I	20	22.7%	47	53.4%	p < 0.001
Stage II	43	48.9%	31	35.2%
Stage III	25	28.4%	10	11.4%
Estrogen Receptor Positive	73	83.0%	76	86.4%	p = 0.676
Estrogen Receptor Negative	15	17.0%	12	13.6%
Progesterone Receptor Positive	66	75.0%	69	78.4%	p = 0.722
Progesterone Receptor Negative	22	25.0%	19	21.6%
HER2 Positive	19	21.6%	16	18.2%	p = 0.574
HER2 Negative	65	73.9%	69	78.4%
HER2 Unknown	4	4.5%	3	3.4%
Bloom-Richardson Grade Low (3, 4, 5)	6	6.8%	21	23.9%	p = 0.007
Bloom-Richardson Grade Intermediate (6, 7)	39	44.3%	37	42.0%
Bloom-Richardson Grade High (8, 9)	37	42.0%	28	31.8%
Bloom-Richardson Grade Unknown	6	6.8%	2	2.3%
Partial Mastectomy	24	27.3%	47	53.4%	p < 0.001
Total Mastectomy	61	69.3%	41	46.6%
No Mastectomy	3	3.4%	0	0%
Radiation Therapy – No	39	44.3	29	33.0%	p = 0.163
Radiation Therapy – Yes	49	55.7%	59	67.0%
Chemotherapy – No	19	21.6%	37	42.0%	p = 0.006
Chemotherapy – Yes	69	78.4%	51	58.0%
Hormone Therapy – No	27	30.7%	16	18.2%	p = 0.079
Hormone Therapy – Yes	61	69.3%	72	81.8%
Total	88	100.0%	88	100.0%

Comparing risk category distribution for recurrent and non-recurrent groups

In response to the first question (describe your current health condition), recurrent patients reported poorer general health than non-recurrent patients (p<0.001, Table 2).

Table 2:

Survey results for general health condition in recurrent and non-recurrent groups (Question 1)

Health Condition	Recurrent group		Non-Recurrent group
	N	%	N	%
Excellent	8	9.1%	10	11.4%
Very Good	31	35.2%	56	63.6%
Good	32	36.4%	21	23.9%
Fair	11	12.5%	1	1.1%
Poor	6	6.8%	0	0.0%
Total	88	100.0%	88	100.0%

Comparing recurrent and non-recurrent groups Fisher’s Exact test = <0.001

In response to the second question (year of first breast cancer diagnosis), 94% of patients accurately identified the date of first diagnosis of breast cancer within 1 year (Table 3). However, the patients with a single breast cancer were more accurate (p<0.001). Patterns of accuracy between recurrent and non-recurrent cohorts were not significantly different (p=0.325, not shown).

Table 3:

Accuracy of self-reported year of first breast cancer diagnosis for women with single and multiple diagnoses (Question 2)

Difference in Years	All Patients		Single Index Breast Cancer		Multiple Breast Cancers
	N	%	N	%	N	%
None	154	87.5%	140	91.5%	14	60.9%
1 year	12	6.8%	7	4.6%	5	21.7%
2 years	3	1.7%	2	1.3%	1	4.3%
3 or more years	7	4.0%	4	2.6%	3	13.0%
Total	176	100.0%	153	100.0%	23	100.0%

^*Comparing single with multiple breast cancer patients Fisher’s Exact test < 0.001

For the third question (whether diagnosed with a new, second breast cancer), 93% (163/176) of the survey responses were correct. The non-recurrent group tended to be more accurate than the recurrent group (97% versus 89%, respectively, p=0.080). Of the 23 patients who correctly identified that they had a second primary breast cancer diagnosed, 4 did not provide a diagnosis year. Of the remainder, 84% (16/19) identified the correct year of diagnosis.

The study participants were then provided the date of the index breast cancer diagnosis and asked the fourth question (identify the index breast cancer initial treatments). The patients correctly identified the various standard medical treatments as follows: surgery, 97.2%; radiation therapy, 95.5%; chemotherapy, 97.7% and hormone therapy, 83.0%. Recurrence of the breast cancer was not related to accuracy (results not shown). Decreased accuracy was seen in multiple breast cancer patients for radiation therapy (single 98.0%, multiple 78.3%, Fisher Exact Test p=0.001, details not shown).

In the fifth question (Did you have return of the breast cancer, when (month/year) and where in your body?), no non-recurrent patients reported a recurrence and 3 recurrent patients failed to report this event (sensitivity 96.6%, specificity 100%, Table 4). In patients reporting recurrence, 89% (76/85) and 79% (67/85) of patients correctly identified the recurrence date within 12 and 6 months, respectively. Patients with multiple breast cancers were slightly less accurate (p=0.082). No significant differences in accuracy were associated with patient age at index cancer diagnosis or year of recurrence. The sites of recurrence were answered correctly 82% (70/85, not shown).

Table 4. Accuracy of Self-Report Survey Answers for Breast Cancer Recurrence, Date of Recurrence, Number of Primary Breast Cancers, and Age at Diagnosis of Index Breast Cancer.

Table 4a.: Accuracy of Survey Answer for Breast Cancer Recurrence

PRO Report	Registry Report
	Recurrent	Non-recurrent
Recurrent	85	0
Non-recurrent	3	88
Total	88	88

Positive Predictive Value 100%, Negative Predictive Value 96.7%, Sensitivity 96.6%, Specificity 100%, Accuracy 98.3%

Case reviews for discrepant survey answers in questions 2–5 identified 3 errors in BCRD in the recurrent and 2 errors in the non-recurrent groups. Answers regarding other cancers (question 6) identified 3 patients who had non-breast cancers after the diagnosis of the index breast cancer (an exclusion criterion not documented in the BCRD). The patient-reports provided an audit of the BCRD and correction of 4.5% (8/176) cases.

Discussion

There is a large literature on the use of patient-reported outcomes to enhance medical decision making and patient care; this body of work focuses on using PRO’s to capture the subjective patient experience, improve patient-clinician communication, symptom management and patient quality of life.²³ In this study, we evaluated whether patients can be queried to provide objective information about recurrence that is adequately reliable for cancer registries. We report on the instrument development, utilization, and accuracy.

We identified recurrent patients and matched non-recurrent patients to them based on age at diagnosis, diagnosis year and total number of primary breast cancers. We did not attempt to match patients for disease characteristics at the time of diagnosis, nor for initial treatment characteristics. It is therefore not surprising that recurrent patients had patterns of higher disease stage, higher tumor grade, and more administration of total mastectomy and chemotherapy.

A challenging aspect of self-report for breast cancer recurrence is the frequency (13%, 23/176) of second primary breast cancers among women who have a breast cancer diagnosis. Both second primaries and recurrence are associated with the cancer “coming back,” and it may be difficult for women to differentiate retrospectively. We asked separate questions regarding each of these but did not ask patients to distinguished between these events as cancer registries routinely record second primary diagnoses. In cases of multiple primaries, questions about recurrence were anchored to the most recent primary diagnosis (index case). Similarly, we excluded patients with a new non-breast primary cancer diagnosis after the index breast cancer as cancer registries routinely record these.

Overall, patients were very accurate (>90%) reporting on the occurrence of the first diagnosis of breast cancer, a subsequent new primary breast cancer and a breast cancer recurrence, and the treatment of the last (index) primary breast cancer. Patients correctly identified their first primary breast cancer date within 1 year (94%), occurrence of a second primary breast cancer (93%), treatment of their last primary breast cancer (94%) and the occurrence of recurrence (98%); with no significant difference between recurrent and non-recurrent groups. We did not explicitly query patients to establish their health literacy or numeracy but expect that accuracy would be lower in populations with lower levels of these metrics.

As might be expected, accuracy regarding specifics such as dates of occurrence and individual treatments tended to drop in patients who had multiple primary cancers (questions 2–5). We anticipated that patient age and the duration from the diagnosis of the index breast cancer would compromise the accuracy of reporting but as others have reported,^24,25 we found no evidence to support this concern. The patients who had a breast cancer recurrence had a little more difficulty distinguishing between a second new breast primary and recurrence (question 3, p=0.080). Given that the recurrent patients did not report being as healthy as the non-recurrent patients and that they had a higher rate of chemotherapy administration initially as well as in an ongoing fashion, the recurrence had little effect on the accuracy of their reporting. Results of this survey demonstrate that the patients have excellent appreciation of the crucial objective events in their cancer journey: cancer diagnosis, cancer recurrence and major cancer treatments. These data suggest that patient self-report could potentially be used to identify or rule out a recurrence event. In turn, registries could focus on a smaller fraction of the prevalent population to collect recurrence data. We²⁶ previously documented cancer registries confronted with new requirements and timelines related to patient care struggle with having incomplete data. The current retrospective study documents that patient survey data can facilitate registry collection of new data requirements (e.g. recurrence).

With any patient-querying instrument, the issue of non-response must be considered. In our survey, the fraction not responding was considerably higher among eligible non-recurrent (63%) than recurrent women (21%). However, we made a much more vigorous attempt to recruit recurrent patients because of the limited number of these potential research subjects. With the excess of potential non-recurrent research subjects, we decreased our phone contact efforts for efficiency of study completion thereby increasing the apparent non-response rate. A slightly higher rate of declining participation in recurrent (17%) than non-recurrent (11%) patients was noted. This may reflect their reported poorer health, ongoing treatment and more aggressive initial therapies (total mastectomy, chemotherapy). It is unknown if patients were queried on an ongoing basis following diagnosis how response rates would change over time. If the association between the likelihood of non-response (missing data) and the outcome of interest (recurrence) persists in this setting with the more vigorous contact approach, this could represent an obstacle to the use of patient self-report for recurrence assessment. Recent incorporation of patient surveys into prospective randomized clinical trials may provide more information regarding the non-response and survey decline issues, and lay the ground-work to explore the value of ongoing patient surveys.^27–29 Scaling this type of patient-facing query for cancer registry use could require using more automated or other electronic approaches, which could also affect response rates.^18,30

Several investigators have combined surveys of ‘objective’ medical care measures with surveys of ‘subjective’ patient reported measures and sometimes made the effort to coordinate terminology.^17,31–38 This is appropriate and to be encouraged. Our results challenge the statement that PROs are only of value for subjective information.³⁸ Our study demonstrates that patients can provide objective information regarding disease, treatment, and outcomes. Patients are willing and capable of being meaningful partners in capturing and utilizing objective data in health care. As stated by Sharma³⁹: “PROs will be used to improve health care delivery, optimize clinical outcomes, monitor treatment progress, and individualize care”. Hopefully, patient surveys will become part of the standard of cancer care from diagnosis, through treatment into follow-up with both subjective and objective data included and available in the medical record for research and population database utilization.

Conclusions

Retrospective patient self-report reliably identified breast cancer recurrence 97% and recurrence date within one year 89%. Patient-reported information should enhance cancer registry data.

Table 4b:

Accuracy of Survey Answers for Date of Recurrence Overall and by Recurrence Year

Months of Difference	All Patients		Recurrence Year 2006–2015		Recurrence Year 2016–2018
	N	%	N	%	N	%
None	22	25.0%	9	19.1%	13	31.7%
<= 6 months	45	51.1%	25	53.2%	20	48.8%
> 6 months to <= 1 year	9	10.2%	5	10.6%	4	9.8%
> 1 year	9	10.2%	7	14.9%	2	4.9%
Reported no recurrence	3	3.4%	1	2.1%	2	4.9%
Total	88	100%	47	100%	41	100%

Comparing Recurrence Years 2006–2015 and 2016–2018 Fisher Exact Test p = 0.402

Table 4c:

Accuracy of Survey Answers for Recurrence Date by Number of Primary Breast Cancers

Months of Difference	Single Breast Cancer		Multiple Breast Cancers
	N	%	N	%
None	21	28.4%	1	7.1%
<= 6 months	38	51.4%	7	50.0%
> 6 months to <= 1 year	7	9.5%	2	14.3%
> 1 year	5	6.8%	4	28.6%
Reported no recurrence	3	4.1%	0	0.0%
Total	74	100%	14	100%

Comparing Single and Multiple Primary Breast Cancers, Fisher’s Exact test = 0.082

Table 4d:

Accuracy of Survey Answers for Recurrence Date by Age < 50 and >= 50 Years at Diagnosis of Index Breast Cancer

Months of Difference	Age < 50		Age >= 50
	N	%	N	%
The same	13	28.3%	9	21.4%
<= 6 months	21	45.7%	24	57.1%
>6 months to <= 1 year	4	8.7%	5	11.9%
> 1 year	6	13.0%	3	7.1%
Reported no recurrence	2	4.3%	1	2.4%
Total	46	100%	42	100%

Comparing Age < 50 and >= 50 Years, Fisher’s Exact test = 0.730