The Association of Delay in Curative Intent Treatment with Survival among Breast Cancer Patients: Findings from the Women’s Health Initiative

Abstract

Purpose:

Delays in adjuvant breast cancer (BC) therapy have been shown to worsen outcomes. However, thus far studies have only evaluated delays to initial treatment, or a particular modality, such as chemotherapy, leaving uncertainty about the role of delay to subsequent therapy and the effects of cumulative delay, on outcomes. We investigated the associations of delays across treatment modalities with survival.

Methods:

We included 3368 women with incident stage I-III BC in the Women’s Health Initiative (WHI) enrolled in fee-for-service Medicare who underwent definitive surgery. This prospective analysis characterized treatment delays by linking WHI study records to Medicare claims. Delays were defined as >8 weeks to surgery, chemotherapy, and radiation from diagnosis or prior treatment. We used Cox proportional hazards models to estimate BC-specific mortality (BCSM) and all-cause mortality (ACM) in relation to treatment delays.

Results:

We found 21.8% of women experienced delay to at least one therapy modality. In adjusted analysis, delay to chemotherapy was associated with a higher risk of BCSM (HR=1.71; 95% CI: 1.07–2.75) and ACM (HR=1.39; 95% CI: 1.02–1.90); delay in radiation increased BCSM risk (HR=1.49; 95% CI: 1.00–2.21) but not ACM risk (HR=1.19; 95% CI: 0.99–1.42). Delays across multiple treatment modalities increased BCSM risk 3-fold (95% CI: 1.51–6.12) and ACM risk 2.3-fold (95% CI: 1.50–3.50).

Conclusions:

A delay to a single treatment modality, and to a greater extent an accumulation of delays were associated with higher BCSM and ACM after BC. Timely care throughout the continuum of breast cancer treatment is important for optimal outcomes.

Introduction

The vast majority of breast cancer cases are treated with the goal of curative intent through surgery, and in appropriate cases adjuvant chemotherapy, radiation therapy, and hormone therapy. Timely receipt of individual treatment modalities has been studied with fairly consistent findings of associations between a delays in treatment and worse breast cancer outcomes, for surgery^1–3, chemotherapy^4–11 and radiation^12–14, but with significant variation in effect size and outcomes among these different treatment modalities.^{1–5,8,12,13,15} For radiation, a paucity of data exists regarding the link between a delay and breast cancer-specific mortality (BCSM). Within clinical practice, there is general consensus that prolonged delays in adjuvant treatment should be avoided due to worse outcomes. Anecdotally, a frequently used timeframe is 12 weeks for initiation of chemotherapy and radiation therapy from last completed modality. For chemotherapy, the data are conflicting with some studies finding worse outcomes only after 12 weeks^4,9,16 and others suggesting that outcomes are linearly related to delays measured from completion of a prior modality^5,8,17 or detrimental with a lesser delay.^10,11 Indeed, there are national quality metrics that set a goal for chemotherapy to be initiated within 4 months from diagnosis and radiation within a year for triple negative breast cancers.¹⁸ These metrics are straightforward to capture in a cancer registry, but are difficult to implement unambiguously in the clinical setting, as they represent coarse guidelines, leaving specialty-specific providers (e.g., medical oncologists and radiation oncologists) without a clear benchmark for optimal time to therapy. Another limitation of prior work is the lack of data in women receiving a series of treatment modalities, as the literature has focused predominantly on individual modalities (surgery, chemotherapy or radiation). These standards, however, could be improved by examining timing of treatment in the context of multiple modalities.

Recently, a linkage was established between Women’s Health Initiative (WHI) data and Medicare claims in a cohort of breast cancer survivors, resulting in a dataset that includes information on adjudicated breast cancer diagnosis, cause of death, detailed treatment data across the spectrum of adjuvant breast cancer care, and socio-demographic factors. We sought to understand the effect of delay of treatment initiation on breast cancer-specific and all-cause mortality across the various treatment modalities for non-metastatic breast cancer patients receiving adjuvant therapy (surgery, radiation, chemotherapy). In particular, we sought to evaluate a lesser delay (i.e., 8 weeks to treatment) than had previously been investigated given data suggestive of linearity and investigated delays to each modality in combination. Such an analysis could inform data-driven guidelines of timeframes for individual treatment modalities.

Methods

Study Population

Between 1993 and 1998, the WHI enrolled 161,808 postmenopausal women ages 50–79 years into an observational study and three clinical trials at 40 clinical centers across the U.S. Eligible women were deemed unlikely to die within three years of enrollment or change residence. Study details have been published previously.^19,20 Participants were initially followed through March 2005 and then were subsequently invited to continue follow-up in two extension studies (2005–2010 and 2010–2020).

During the main WHI study period, cancer diagnoses were updated annually in the observational study or semiannually in the clinical trials, and were updated annually for all women participating in the extension studies. Clinical information regarding breast cancer characteristics, including staging (according to Surveillance, Epidemiology, and End Results [SEER] staging), estrogen receptor (ER), progesterone receptor (PR), Her2neu receptor (HER2) expression, was based on pathology review by centrally trained adjudicators.^19,20 Data from women enrolled in the WHI were linked to Medicare enrollment and claims data through December 2014 from the Centers for Medicare and Medicaid Services (CMS) by social security number, birth or death date (or partial date), or zip code.

For the present study, we included WHI participants with a confirmed diagnosis of stage I-III breast cancer, adjudicated through the WHI, from January 1, 1996 through December 31, 2013, who were ≥65 years old at diagnosis, and were enrolled in fee-for-service Medicare A+B at the time of diagnosis. Among the 3,785 women who met these criteria, we excluded those with a pre-existing malignancy (other than non-melanoma skin cancers) or a diagnosis of second invasive, non-breast primary cancer within 4 months of their breast cancer diagnosis as these were potentially synchronous cancers (n=24). We then excluded those without a full year of coverage in Medicare post-diagnosis or who died within the first year after diagnosis (n=242) because either they lacked information on treatment (e.g., resulting from switching to different insurance) or did not survive long enough to complete treatment. We excluded women who received neoadjuvant therapy due to the small number of cases (n=56). We excluded women who did not have cancer-directed surgery as part of their treatment (n=92). Lastly, we excluded women with no claims in the year prior to diagnosis (n=3), resulting in 3,368 women in our analysis.

Treatment delays

We defined a delay as >8 weeks (56 days) of time between any two treatment modalities (Figure 1). We calculated time to treatment for initial breast cancer treatments (surgery, chemotherapy and radiation) using cancer diagnosis date and days between treatment billing dates. We used established Healthcare Common Procedure Coding System (HCPCS) and International Classification of Diseases-9 (ICD9) codes. We then determined the days from the end of one treatment modality to the beginning of the next modality. For example, for women receiving radiation, the number of days was computed between the start date of radiation and the end date of chemotherapy, if she received chemotherapy prior, or if not, from the date of the last surgery. In order to guard against a recurrence or second primary being the source of the treatment, we looked for 4+ month gaps in chemotherapy (which was considered a restart of chemotherapy), and only counted chemotherapy administered before any 4-month gap as relevant to the assessment of treatment.

Figure 1. The conceptual framework modeling timeframe between treatment modalities.

The schematic presents the timeframe for three typical treatment scenarios over time for (A) a patient treated with chemotherapy and radiation after surgery; (B) a patient treated with chemotherapy after surgery; and (C) a patient treated with radiation. The dashed lines represent the time between treatment modalities for which delays to a treatment modality were assessed.

Our main analysis investigated the effect of delays >8 weeks on treatment. The selection of 8 weeks for the timeframe of delay was selected to balance the results from meta-analysis studies demonstrating that a delay of 4 weeks was associated with reduced survival^5,8,17 and multiple other reports that defined a delay at 12 weeks.^4,9,16 Across each treatment modality, the extent of outliers was minimal (i.e., <10 across each modality). As a sensitivity analysis, we created a continuous variable for weeks between individual treatments, as well as categorical variables for each 4-week period of time to initiation of the next treatment (in which 4–8 weeks served as the reference category, as data were sparse for <4 weeks). For cumulative delay, we summed across the treatment modalities using the dichotomous treatment delay variable (</>8 weeks).

Mortality

Outcomes included BCSM and all-cause mortality (ACM). Participants were followed for these outcomes from breast cancer diagnosis until February 28, 2017. Determination of a death was determined through linkage to the National Death Index (NDI), as well as through WHI study procedures, such as proxy informant’s response to the WHI annual survey and WHI local sites’ monitoring of obituary notices. Cause of death was determined through NDI records, followed by medical record review and central adjudication at the WHI clinical coordinating center.

Covariates

On standardized questionnaires at baseline, women provided information on age, race/ethnicity (Non-Hispanic White, Black, and other non-White), attained education (less than or equivalent to high school diploma/GED; schooling after high school; college degree or higher), yearly family income (<$20,000; $20,000–$49,999; $50,000–$74,999; ≥$75,000), smoking status (never/ever), US region of residence (Northeast, South, Midwest, and West). We categorized rural-urban residences using the Rural Urban Commuting Area (RUCA) codes based on zip code at diagnosis. Body-mass index (BMI, kg/m2) was calculated from weight and height measured at the baseline clinic visit, and categorized as normal weight, overweight, and obese. Comorbidity was determined from Medicare data for the year preceding the diagnosis of breast cancer using the modified Charlson score. When unknown values were present for the variables listed above, a category of unknown values was included in the modeling (except when noted).

Statistical Analysis

We evaluated treatment delays (>8 weeks) in relation to demographic and tumor characteristics using t-tests (for continuous variables), Cochran-Armitage tests of linear trend (for ordinal variables), or Chi-square tests (for categorical variables). Of note, we did not show individual rows of data within our tables for cell size <11, according to CMS requirements.

Hazard ratios (HR) and 95% confidence intervals (CI) for BCSM and ACM in relation to time to surgery, time to chemotherapy, time to radiation, and cumulative delays were estimated by Cox proportional hazard models. BCSM was defined as time from one-year post-breast cancer diagnosis to death due to breast cancer, and event times were censored at the time of a non-breast cancer related death, or at last documented follow-up. ACM was similarly defined as the time from one-year post-breast cancer diagnosis to any death, and event times were censored at the last documented follow-up. Follow-up time began one year post-diagnosis because women were required to be enrolled and alive through one-year post-diagnosis, and the exposures of interest took place during this year. For each category of delay, the analysis was restricted to women who had received that therapy. Likewise, the analysis of cumulative delays was restricted to those who had received at least 2 treatment modalities. Models were adjusted for age at diagnosis; year of diagnosis; race/ethnicity; tumor size (≤1 cm, >1–2, >2–5, >5); nodal involvement; ER, PR, and HER2 expression (each defined as positive, negative, or unknown); and Charlson comorbidity score (0, 1, 2+). Linear trends were evaluated by testing the integer-scored variable as a continuous variable in the models. We tested inclusion of US region, rural-urban residence (via RUCA codes)¹ at the time of cancer diagnosis, smoking status, diabetes, and baseline categorization of education, family income, and BMI, to the models; none made changes of 10% or more to the hazard ratios, and thus were not retained in the models.

Effects of treatment delays and completion by race/ethnicity and age at diagnosis (<75, 75+) subgroups were assessed by including interaction terms to test multiplicative interactions in the models. Analyses were conducted using SAS version 9.4 (SAS Institute Inc., Cary, NC, USA). P-values were based on two-sided tests and considered significant at 0.05.

Results

In the cohort of 3,368 breast cancer patients, women were followed for an average of 9.7 years. On average, women were 74 years of age, 90% were non-Hispanic White, and 44% had a 4-year college degree or higher (Table 1). With respect to breast cancer characteristics, 20% had nodal involvement, 83% were ER+, 10% were HER2+, and 7% were triple negative. In this cohort, 68% had breast-conserving surgery (BCS); 64% received radiation and 19% received chemotherapy. Of those receiving BCS, 84% received radiation; 20% of women who had mastectomies received radiation. In our cohort 1,085 deaths occurred; of those, 883 were breast cancer-specific.

Table 1:

Characteristics of the study sample

	Breast cancer cases (n=3368)
	N	%
Demographics1
Age at cancer diagnosis
65–69	840	24.9
70–74	1035	30.7
75–79	899	26.7
≥ 80	594	17.6
Race/ethnicity
Non-Hispanic White	3045	90.4
African American or Black	180	5.3
Non-White/non-Black	143	4.2
Residence at diagnosis, percent urban
0	154	4.8
1–49	218	6.8
50–99	614	19.2
100	2210	69.1
Education
High school diploma/GED or less	626	18.7
School after high school	1232	36.8
College degree or higher	1492	44.5
Body-mass index( kg/m2)
<25	1109	33.2
25 – <30	1147	34.3
>=30	1087	32.5
Smoking status
Never	1644	49.5
Past	1516	45.7
Current	158	4.8
Charlson comorbidity score3
0	2482	73.7
1	611	18.1
2+	275	8.2
Tumor characteristics
Summary Stage
Localized	2669	79.2
Regional	699	20.8
Tumor size (cm)
≤1	1358	40.3
>1–2	1285	38.2
>2–5	565	16.8
>5	64	1.9
Unknown	96	2.9
Nodal involvement
No	2667	79.2
Yes	662	19.7
Unknown	39	1.2
ER receptor
Positive	2787	82.7
Negative	426	12.6
Unknown	155	4.6
PR receptor
Positive	2399	71.2
Negative	795	23.6
Unknown	174	5.2
Her2-neu receptor
Positive	345	10.2
Negative	2367	70.3
Unknown	656	19.5
Triple negative
No	3131	93.0
Yes	237	7.0
Cancer treatment2
Chemotherapy
No	2724	80.9
Yes	644	19.1
Surgery type and radiation
Breast-conserving with no radiation	361	10.7
Breast conserving with radiation	1945	57.8
Mastectomy with no radiation	852	25.3
Mastectomy with radiation	210	6.2

¹Demographic characteristics as collected at enrollment in the Women’s Health Initiative, except as noted.

²Derived from Medicare claims data

Delay to surgery >8 weeks occurred for 4.2% of the women and was associated with increasing age, race (6.5% for non-White women, p=0.03) and increasing comorbidities (10.9% for Charlson score of 2+, p<0.001) (Table 2). A delay to chemotherapy of > 8 weeks occurred in 23.0% of women, and was disproportionately experienced by non-White women, in whom 35.0% experienced a delay (p=0.006). Notably, localized disease was also associated with delay to chemotherapy (27.5%, p=0.02). A delay to radiation >8 weeks from prior treatment occurred in 22.5% of women, and was associated with increasing age (p=0.004) and increasing Charlson comorbidities score (p<0.001).

Table 2:

Treatment delays by characteristics of the study sample^1,2

	Surgery			Chemotherapy			Radiation Therapy
	Received surgery (n=3368)	Delay >8 weeks (n=143)	P-value3	Received chemotherapy (n=644)	Delay >8 weeks (n=148)	P-value3	Received radiation (n=2155)	Delay >8 weeks (n=484)	P-value3
	N	N (%)		N	N (%)		N	N (%)
Age at diagnosis
65–69	840	22 (2.6)	<0.001	251	52 (20.7)	0.05	574	110 (19.2)	0.004
70–74	1035	38 (3.7)		223	47 (21.1)		708	157 (22.2)
75–79	899	48 (5.3)		133	37 (27.8)		569	133 (23.4)
80+	594	35 (5.9)		37	12 (32.4)		304	84 (27.6)
Race/ethnicity
White	3045	122 (4.0)	0.03	564	120 (21.3)	0.006	1950	437 (22.4)	0.87
Non-White	323	21 (6.5)		80	28 (35.0)		205	47 (22.9)
Education
High school diploma/GED or less	626	22 (3.5)	0.88	124	29 (23.4)	0.40	391	81 (20.7)	0.07
School after high school	1232	60 (4.9)		235	60 (25.5)		791	164 (20.7)
College degree or higher	1492	60 (4.0)		279	58 (20.8)		964	236 (24.5)
Charlson comorbidity score
0	2482	83 (3.3)	<0.001	490	114 (23.3)	0.75	1635	346 (21.2)	<0.001
1	611	30 (4.9)		108	20 (18.5)		370	87 (23.5)
2+	275	30 (10.9)		46	14 (30.4)		150	51 (34.0)
Summary stage
Localized	2669	107 (4.0)	0.18	276	76 (27.5)	0.02	1684	368 (21.9)	0.20
Regional	699	36 (5.2)		368	72 (19.6)		471	116 (24.6)
Tumor size (cm)
≤1	1358	51 (3.8)	0.02	87	24 (27.6)	0.05	913	197 (21.6)	0.20
>1–2	1285	46 (3.6)		265	67 (25.3)		829	185 (22.3)
>2–5	565	37 (6.5)		230	48 (20.9)		308	77 (25.0)
>5	64	*		41	*		42	11 (26.2)
Nodal involvement
Yes	662	32 (4.8)	0.42	357	69 (19.3)	0.01	452	112 (24.8)	0.17
No	2667	110 (4.1)		284	78 (27.5)		1675	364 (21.7)
Estrogen receptor
Positive	2787	122 (4.4)	0.55	412	91 (22.1)	0.41	1816	406 (22.4)	0.94
Negative	426	16 (3.8)		195	49 (25.1)		253	56 (25.6)
Progesterone receptor
Positive	2399	99 (4.1)	0.43	336	76 (22.6)	0.77	1553	353 (22.7)	0.43
Negative	795	38 (4.8)		271	64 (23.6)		508	107 (21.1)
Her2-neu receptor
Positive	345	21 (6.1)	0.29	112	25 (22.3)	0.76	213	44 (20.7)	0.53
Negative	2367	113 (4.8)		424	89 (21.0)		1537	347 (22.6)

^1.Numbers do not always add to the column total, due to missing data.

^2.Numbers for cells with fewer than 11 participants are not shown.

^3.P-values are from chi-square tests of association or Cochran-Armitage test for a linear trend in proportions (age at diagnosis, education, Charlson comorbidity score, and tumor size).

When examining individual treatment modalities, delay to chemotherapy was associated with a higher risk of BCSM and ACM (Table 3). We found that a delay >8 weeks was associated with a 71% higher risk of BCSM (95% CI: 1.07–2.75). With respect to ACM, chemotherapy delay was associated with a 1.4-fold elevated risk (95% CI: 1.02–1.90). In the assessment of time to treatment, each additional week to chemotherapy was associated with an 8% (95% CI: 1.04–1.11) higher risk of BCSM after adjustment for age, year of diagnosis, race/ethnicity, breast cancer characteristics, and Charlson comorbidity score.

Table 3.

Risk of Breast Cancer Specific Mortality and All-Cause Mortality associated with Time to Treatment across Individual Treatment Modalities

		Breast Cancer Mortality4			All-Cause Mortality4
	Total N (%)	N	Hazard Ratio	95% CI	N	Hazard Ratio	95% CI
Delay to treatment1 (modeled categorically, >8 weeks)
Delay to chemotherapy2
No	496 (77.0)	61	1.00	Reference	152	1.00	Reference
Yes	148 (23.0)	28	1.71	(1.07, 2.75)	63	1.39	(1.02, 1.90)
p-value				0.03			0.04
Delay to radiation3
No	1671 (77.1)	89	1.00	Reference	480	1.00
Yes	484 (22.9)	36	1.49	(1.00, 2.21)	162	1.19	(0.99, 1.42)
p-value				0.047			0.06
Time to treatment (modeled continuously, by weeks)
Time to surgery
From diagnosis to surgery	3368		0.97	(0.91, 1.02)		1.00	(0.98, 1.02)
p-value				0.26			0.85
Time to chemotherapy2
From surgery to chemotherapy	644		1.08	(1.04, 1.11)		1.04	(1.02, 1.07)
p-value				<0.0001			0.0002
Time to radiation3
From prior treatment to radiation	2155		1.02	(0.98, 1.06)		1.02	(0.998, 1.03)
p-value				0.45			0.08

^1.As the majority of categories of delay (>8 weeks) for surgery had cell size < 11, data are not shown; no relationships were statistically significant.

^2.Chemotherapy delay was only analyzed for women who had chemotherapy; radiation delay was only analyzed for women who had received radiation.

^3.From chemotherapy to radiation; or if no chemotherapy received, then from surgery.

^4.Each model was adjusted for age at diagnosis (continuous), year of diagnosis (continuous), race/ethnicity, tumor size, nodal involvement (yes/no), ER (positive, negative, unknown), PR (positive, negative, unknown), Her2-neu receptor (positive, negative, unknown), and Charlson comorbidity score (0, 1, 2+).

In sensitivity analysis, we found the increased risk of BCSM in relation to a delay >12 weeks for chemotherapy versus 4–8 weeks (the most numerous category) was 3.28 (95% CI: 1.73–6.24; data not shown). Women who encountered a delay to chemotherapy >12 weeks were similar to those with an 8-week delay across socio-demographic factors and tumor characteristics, with the possible exception of race, as 15.0% of non-White women had >12 week delay versus 6.9% of White women (p=0.05). Of those with a delay >12 weeks, BCSM occurred in a higher percentage (42.9%) of women with less than a high school education, compared to the women who survived breast cancer (13.5%; p=0.04).

A delay to radiation >8 weeks was associated with an increased BCSM risk (HR=1.49; 95% CI: 1.00–2.21; p=0.047). We did not find an increase in BCSM or ACM when looking at time to radiation therapy using a continuous variable (HR=1.02; 95% CI: 0.98–1.06), nor in ACM risk in relation to 8-week delay (HR=1.19; 95% CI: 0.99–1.42). In the sensitivity analysis investigating time between treatment by 4-week increments, we observed that a delay >8–12 weeks to radiation was associated with a 1.8-fold risk of BCSM (95% CI: 1.13–3.01), while a delay >12 weeks was not significantly associated with BCSM. Likewise, for ACM, a delay of >8–12 weeks was associated with a 1.2-fold increased risk of ACM (95% CI: 1.01–1.54), while a >12 week delay was not (data not shown). There was no indication that delay to surgery was related to BCSM or ACM, although a delay to surgery was uncommon, and thus our data were likely underpowered in this analysis.

When cumulative treatment delays were considered, 2 or more delays (of >8-weeks) across treatment modalities were associated with a 3.0-fold increased BCSM risk (95% CI: 1.51–6.12) and a 2.3-fold higher ACM risk (95% CI:1.50–3.50) after controlling for tumor characteristics and comorbidities (Table 4). This was comparable to the risk of BCSM for nodal involvement at 3.05 (CI: 2.16–4.31). Neither age nor non-White race modified the effect of treatment delay on BCSM or ACM.

Table 4:

Risk of Breast Cancer Specific Mortality and All-Cause Mortality associated with Cumulative Treatment Delays, with Results Presented for a Multi-variate Model^1,2

	Breast cancer Mortality			All-Cause Mortality
	HR	95% CI	P-value	HR	95% CI	P-value
Age at diagnosis	1.07	(1.03,1.11)	<0.0001	1.13	(1.11,1.14)	<0.0001
Year of diagnosis	0.86	(0.82,0.91)	<0.0001	0.96	(0.94,0.99)	0.008
Race/ethnicity
Non-Hispanic white	1.00	Reference	1.00	1.00	Reference	0.23
Non-Hispanic Black	1.00	(0.53,1.86)		0.81	(0.57,1.14)
Non-White/non-Black	1.02	(0.41,2.52)		0.76	(0.49,1.17)
Tumor size (cm)
≤1	1.00	Reference	<0.0001*	1.00	Reference	0.0002*
>1–2	2.33	(1.40,3.85)		1.15	(0.97,1.38)
>2–5	3.51	(2.03,6.05)		1.43	(1.14,1.78)
>5	5.01	(2.32,10.81)		1.91	(1.24,2.94)
Unknown	3.47	(1.46,8.24)		1.21	(0.79,1.84)
Nodal involvement
No	1.00	Reference	<0.0001**	1.00	Reference	0.0003**
Yes	3.05	(2.16,4.31)		1.38	(1.16,1.65)
Unknown	0.66	(0.09,4.87)		1.12	(0.64,1.97)
Estrogen receptor
Positive	1.00	Reference	0.0005**	1.00	Reference	0.04**
Negative	2.20	(1.38,3.53)		1.29	(1.01,1.65)
Unknown	1.01	(0.17,6.05)		0.88	(0.38,2.07)
Progesterone receptor
Positive	1.00	Reference	0.15**	1.00	Reference	0.39**
Negative	1.37	(0.89,2.11)		1.09	(0.89,1.33)
Unknown	1.34	(0.25,7.26)		1.05	(0.47,2.35)
Her2-neu receptor
Positive	1.00	Reference	0.9**5	1.00	Reference	0.56**
Negative	1.02	(0.63,1.66)		0.94	(0.74,1.20)
Unknown	1.02	(0.57,1.79)		1.06	(0.80,1.40)
Charlson comorbidity score
0	1.00	Reference	0.002*	1.00	Reference	<0.0001*
1	1.41	(0.93,2.14)		1.70	(1.41,2.05)
2+	2.42	(1.34,4.37)		3.10	(2.42,3.98)
Cumulative delays
0	1.00	Reference	0.013*	1.00	Reference	0.004*
1	1.23	(0.87, 1.74)		1.13	(0.96, 1.33)
2–3	3.04	(1.51, 6.12)		2.29	(1.50, 3.50)

^1.Adjusted for all variables shown, including age at diagnosis (continuous), race/ethnicity, tumor size, nodal involvement, ER status, PR status, Her2-neu receptor status, Charlson comorbidity score, as well as year of diagnosis.

^2.Restricted to women who received ≥2 treatment modalities as that was a requirement for the analysis of cumulative delays.

^*trend p-value

^**p-value excludes unknown category

Discussion

In this analysis of breast cancer patients 65 years and older from the WHI, we found that 21.8% of women overall had a delay >8 weeks for at least one modality of therapy. A delay of 8 weeks to chemotherapy was associated with a 1.7-fold higher mortality from breast cancer and a 1.4-fold higher risk of ACM; a delay of 8 weeks to radiation increased BCSM by 1.5. Furthermore, multiple delays across the modalities of breast cancer treatment increased risk of breast cancer mortality by 3-fold.

Prior research has primarily focused on the effect of delay to a single treatment modality, such as surgery, chemotherapy, radiation, or initial treatment.^1–14,21 Of these, the association of survival with delays to adjuvant chemotherapy has been the predominant focus. In several meta-analyses, the time to chemotherapy was found to increase risk of death by 4–15% for each 4-week delay.^8,15,17 The results of our analysis, an 8% risk per week from surgery to chemotherapy is in broad agreement with prior results seen in clinical trial and observational data. In our adjusted analysis, we demonstrated that a delay to chemotherapy >8 weeks was associated with both a higher risk of BCSM (HR=1.71; 95% CI: 1.07–2.75) and ACM (HR=1.39; 95% CI: 1.02–1.90) in a population that was primarily hormone receptor positive. Notably, we found a higher risk of delay to chemotherapy in those with localized versus regional disease. Similar findings have been seen^6,7 and partially attributed to tests such as oncotypeDx that assist in the decision to use chemotherapy but may also lead to a risk of delay to chemotherapy. Broader recognition, quality metrics and systems processes may help to decrease delays.

The effect of delay to radiation on breast cancer outcomes has been less well studied. The effects of radiation on survival are difficult to detect, as this is usually mediated by changes in local recurrence. Prior studies have demonstrated a higher risk of local recurrence over time in relation to radiation delay (HR 1.08/month¹³; 1.005/day¹⁴) after breast conserving surgery in those not receiving chemotherapy.^13,14,22 With respect to survival, the effect of delayed radiation has been more variable. In a meta-analysis primarily including randomized trials evaluating sequencing of chemotherapy and radiation^23–25 and excluding registry studies, no effect was seen on survival with delay of radiation (HR 0.99; 95% CI: 0.94–1.05); however when several large registry studies¹² were included they found the risk of death was 1.06 for every additional month to treatment (95% CI: 1.04–1.07).¹³ In a large retrospective study of women with Stage I-II breast cancer not receiving radiation, there was decreased survival only for >12 weeks prior to radiation initiation.¹² There is exceedingly little information available on delay to post-mastectomy radiation with one study not finding an effect on survival in a cohort of 340 women. However, radiation delay was defined from the time of diagnosis, with no consideration of surgery or chemotherapy.¹⁵ In summary, very few studies have evaluated time to radiation after chemotherapy, and these evaluated time from diagnosis to radiation rather than evaluating delay between treatments. However, this approach conflates delays across modalities, as it would be unclear if any increased risk was due to delay in chemotherapy or radiation. A preferable approach is to investigate the time between treatment modalities, as this more accurately identifies when delay occurs and thus is more clinically actionable. Our analysis used the latter approach, finding that a delay in radiation after prior treatment (either surgery or chemotherapy) was associated with BCSM (HR=1.49; 95% CI: 1.01–2.22). The discrepancy between our finding and prior studies may be that we included women with higher risk disease (as we included stage III and those receiving chemotherapy, whom were excluded in other analyses). In this setting it is conceivable that the increased BCSM risk observed in relation to radiation acts through a pathway involving a recurrence or second primary which our study did not directly interrogate, but which is feasible within the 9.7 years of follow-up on average in our study. In this analysis we did not evaluate the appropriateness of radiation therapy, but rather the association of time to radiation, if administered, with survival.

Our study adds to the literature by evaluating the association of cumulative delays with BCSM and ACM. Despite the small percentage of women in our study with multiple delays (1.2% and 2.6% of those who received 2 modalities and 3 modalities, respectively), we observed that multiple delays had a similar effect size with BCSM as did tumor characteristics and co-morbidities. Importantly, delay is a modifiable factor which could be targeted by process interventions, in contrast to both comorbidities and tumor characteristics. However, we must acknowledge that this study cannot establish causality between treatment delays and mortality, as it cannot rule out that confounding played a role in the associations. It is possible that the reasons underlying a delay have impacted survival. We were, however, able to adjust for a comprehensive set of factors in this well-characterized patient population, and observed that the relationship between cumulative delay and mortality persisted after controlling for measured confounding factors.

Breast cancer care is multidisciplinary and complicated with many possible barriers and causes of delay.^2,3,21 Prior studies have predominantly focused on delay to a particular breast cancer therapy. These studies have found that treatment delays have been associated with increasing age, comorbidities, and African American race.^1–4 In our data, we also observed risk factors for treatment delays included age, co-morbidities, and for delay to surgery and chemotherapy, non-White race. Our finding that the delay in treatment modality for non-white women did not occur for radiation is similar to a single institution study that found a delay to surgery but not to subsequent treatments was associated with race/ethnicity.²⁶ Hershman et al, however, found that Black race was associated with both delays in chemotherapy and radiation in the SEER-Medicare population.⁴ This suggests that there may be factors that moderate the risk of delay for Black women diagnosed with breast cancer.

Our study has several strengths. Its comprehensive evaluation of treatment timing across multiple modalities was made possible by the linkage between adjudicated breast cancer outcomes in the WHI and Medicare treatment data, as well as specifying the cause of death due to breast cancer. We also acknowledge the limitations of our study which includes that our analysis was limited to women receiving Medicare fee-for-service, and as such did not include Medicare Advantage plans. This limits the generalizability of our study findings; however, this also controls for the potential influence of differing plans, such that the health care plan was the same across the study population. An additional limitation was the inability to evaluate delay or utilization to hormone therapy in this dataset (due to the difference in availability from claims data). Similarly, we were not able to investigate the role of oncotypeDX in delays to chemotherapy associated with localized stage, as the billing code for this test was non-specific during much of the study timeframe. We did not correct for multiple testing; however, these analyses were all hypothesis driven such that our statistical tests were designated a priori. Our power to detect particular associations was limited within some subgroups. Lastly, these findings must be interpreted in the context of the study population in which a large proportion of women are college-educated, which may extrapolate to medical literacy. As such, these findings may not be generalizable to all populations of women diagnosed with breast cancer.

In summary, we found that delays in chemotherapy greater than 8 weeks and multiple delays across treatments were associated with higher breast cancer-specific and all-cause mortality. We also observed an association between a delay to radiation >8 weeks and increased BCSM. Effect sizes for multiple delays and BCSM (3-fold) were on par with cancer characteristics, such as nodal involvement (3-fold). Minimal national quality metrics exist (i.e. chemotherapy within 4 months of diagnosis and radiation therapy within 1 year of diagnosis), but these may not be stringent enough to offset the worse outcomes seen with delays. Additionally, these metrics and most studies have focused on delay from diagnosis to a therapy, which is easy to calculate from a registry but are not generally clinically actionable. Our data suggest that cumulative delays between individual treatments is associated with increased mortality. If future studies replicate these findings, this would indicate a clinical focus on reducing delays between individual treatments could be an actionable goal to improve survival in women with breast cancer.