KRAS Testing and Epidermal Growth Factor Receptor Inhibitor Treatment for Colorectal Cancer in Community Settings

Abstract

Background

In metastatic colorectal cancer (mCRC), mutations in the KRAS gene predict poor response to epidermal growth factor receptor (EGFR) inhibitors. Clinical treatment guidelines now recommend KRAS testing if EGFR inhibitors are considered. Our study investigates the clinical uptake and utilization of KRAS testing.

Methods

We included 1,188 patients with mCRC diagnosed from 2004 to 2009, from seven integrated health care delivery systems with a combined membership of 5.5 million. We used electronic medical records and targeted manual chart review to capture the complexity and breadth of real-world clinical oncology care.

Results

Overall, 428 patients (36%) received KRAS testing during their clinical care, and 266 (22%) were treated with EGFR inhibitors. Age at diagnosis (p=0.0034), comorbid conditions (p=0.0316), and survival time from diagnosis (p<0.0001) influence KRAS testing and EGFR inhibitor prescribing. The proportion who received KRAS testing increased from 7% to 97% for those treated in 2006 and 2010, respectively, and 83% of all treated patients had a KRAS wild type genotype. Most patients with a KRAS mutation (86%) were not treated with EGFR inhibitors. The interval between mCRC diagnosis and receipt of KRAS testing decreased from 26 months (2006) to 10 months (2009).

Conclusions

These findings demonstrate rapid uptake and incorporation of this predictive biomarker into clinical oncology care.

Impact

In this delivery setting, KRAS testing is widely used to guide treatment decisions with EGFR inhibitors in patients with mCRC. An important future research goal is to evaluate utilization of KRAS testing in other delivery settings in the US.

Introduction

KRAS testing is used to help make treatment decisions for patients with metastatic colorectal cancer (mCRC). The KRAS gene is present in tumors in two forms: mutated and wild-type. For patients whose tumor tissue expresses the wild-type KRAS genotype, combination treatment with epidermal growth factor receptor (EGFR) inhibitors and chemotherapy has been shown to improve survival (1). Patients with the mutated form of KRAS do not experience this survival benefit. Thus KRAS testing allows oncologists to tailor the use of EGFR inhibitors, cetuximab (Erbitux®, ImClone Systems Incorporated, New York, NY) or panitumumab (Vectibix®, Amgen Incorporated, Thousand Oaks, CA), to increase treatment effectiveness, minimize adverse events, and be cost effective.

In February 2009, the American Society of Clinical Oncology (ASCO) recommended that “All patients with mCRC who are candidates for anti-EGFR antibody therapy should have their tumor tested for KRAS mutations” (2). The National Comprehensive Cancer Network (NCCN) guidelines were revised in November, 2008 to recommend EGFR inhibitors only for patients with KRAS wild-type genotype (3). This was revised again to include cetuximab and panitumumab as first line therapies in 2009 and 2011, respectively (4, 5). The FDA also changed labeling for EGFR inhibitors to describe the appropriate use of KRAS genetic testing (6). No studies have yet examined how KRAS testing has been disseminated in general practice in the U.S. This study addresses this gap and is among the first to assess characteristics associated with KRAS testing across multiple integrated health care delivery systems serving diverse communities.

In this study, we examine factors previously associated with variable adoption of technologies for cancer diagnosis and treatment, such as advanced age, poor pre-treatment health status, minority race-ethnicity, lower socioeconomic status, and higher comorbidity. Because EGFR inhibitors were recommended primarily as second-line therapies during the study period, we examined whether patient factors are associated with KRAS testing. We describe real-world trends in adoption of KRAS testing, timing of KRAS testing relative to cancer diagnosis and chemotherapy initiation, use of EGFR inhibitors by KRAS test status and result, and variations in testing and treatment across study sites. The overall purpose of these analyses is to help guide future efforts to disseminate other novel genomic tests.

Methods

Research Environment

This research was part of the Comparative Effectiveness Research in Genomics of Colon Cancer (CERGEN) study, which includes investigators from eight Cancer Research Network (CRN) sites and partners from academic institutions (7). We collected data at seven CRN sites across the U.S. representing diverse populations. Integrated health care systems have: 1) a defined population; 2) capitation payment; 3) ownership of medical offices, hospitals, and pharmacies; 4) an integrated medical record; and 5) exclusive relationships with one or more medical groups. Although not all integrated health care systems include all of these components, the key concept is that the health plan faces a single global budget which must pay for all medical care services. In 2008, about 25% of Americans received healthcare in Health Maintenance Organizations (8).

Definition of the Eligible Patient Population

The study population includes 4,446 patients enrolled at one of seven CRN study sites: Kaiser Permanente Northwest (Oregon and Washington), Kaiser Permanente Northern California, Kaiser Permanente Colorado, Kaiser Permanente Hawaii, Marshfield Clinic (Wisconsin), Henry Ford Health System (Michigan), and HealthPartners (Minnesota and Western Wisconsin). Eligible patients were identified through tumor registries linked with electronic health information.

Eligible cases were those initially diagnosed with stage IV CRC between January 1, 2006 and December 31, 2009, and patients initially diagnosed with stage III CRC between January 1, 2004 and December 31, 2006, according to the criteria of the American Joint Committee on Cancer (AJCC) (9). The earlier period for the stage III CRC cases allows adequate follow-up time to determine whether those cases eventually progressed to distant metastatic disease. Patients initially diagnosed at stage III were excluded if they did not progress to distant metastatic CRC. Patients initially diagnosed at stage III were included to increase the available sample size, and to evaluate whether KRAS test utilization differs depending on the initial stage at diagnosis. To ensure ascertainment of complete treatment data, we included only cases diagnosed while affiliated with the integrated health care delivery systems. Patients were excluded if they were diagnosed under the age of 18 years or if they did not maintain health plan affiliation for at least one year following diagnosis, allowing for a three month gap in affiliation. Patients with less than one year of follow-up were not excluded if the reason for disaffiliation was death. Patients were censored for death at any time after diagnosis or disaffiliation after 12 months following diagnosis.

This study was approved by the Institutional Review Boards (IRB) at Kaiser Permanente Northwest, Kaiser Permanente Hawaii, Kaiser Permanente Colorado, Marshfield Clinic Research Foundation, and Henry Ford Health System, and did not require written informed consent. The IRBs for the remaining sites ceded authority to the Kaiser Permanente Northwest IRB. A small number of members at each health plan have elected not to participate in anonymous or unconsented research protocols; thus, these patients were excluded.

Data Collection

Each CRN site maintains electronic databases in a common, shared format called the virtual data warehouse (VDW) (10), enabling us to use distributed code to extract clinical data on each participant. We used the vital signs, procedures, pharmacy, enrollment, encounter, diagnosis, and census databases. Each CRN site also maintains a tumor registry where clinical data are abstracted from the medical chart into an electronic database. We used the VDW tumor registry files to identify eligible cases and obtain electronic data. We queried electronic databases for patient characteristics [e.g., gender, age at diagnosis, race/ethnicity], tumor characteristics (cancer site, stage, histology), and treatment history (chemotherapies used and indicators of immunotherapy, radiation treatment, surgical treatment, hormone treatment, palliative care). We used group level measures for socio-economic status (SES) derived from geo-coding (median household income and proportion of individuals in a census tract with a high school education), since individual level factors were not available. The Charlson comorbidity index was computed based on diagnosis information (11).

KRAS Genetic Testing

We used the KRAS genetic test data that was part of routine medical care. We obtained test information either by contacting each site's commercial vendor(s) for KRAS testing directly and asking for a report, or through chart review. Data were abstracted from individual reports including the date of testing, ordering physician, and test result. We have described the methods used for genotyping and assessing the comparability in test results across testing laboratories (12).

Chart Abstraction

Abstractors at each site manually extracted information on each study subject using standard data collection forms. Abstracted variables included verification of eligibility, demographics (race, ethnicity, smoking, alcohol use), family history of cancer, cancer treatment history including surgery, radiation, and chemotherapy, palliative care, genetic testing (including KRAS), and imaging to assess disease progression.

We trained abstractors by developing instructions and an abstraction manual, and held a web-based training session. All abstraction forms were reviewed by a second reviewer to evaluate completeness and case eligibility. Abstraction forms were entered into an electronic database using double data entry to ensure accuracy.

Statistical Analysis

Logistic regression modeling with KRAS testing as the outcome was conducted to identify factors that influence whether a patient was tested. Logistic regression was also used to evaluate factors affecting whether KRAS wild-type subjects were treated with EGFR inhibitors, using a binary variable indicating any use of cetuximab or panitumumab as the outcome of interest, and the covariates described below. Nonparametric analysis of variance was used to characterize differences in interval to testing, since test interval was not normally distributed. All reported p-values are for univariate models (unless otherwise stated), and are unadjusted for multiple testing. All confidence intervals reported from logistic regression are adjusted for the number of factors in the model using a Bonferroni correction.

KRAS testing was coded as a binary indicator, with a value of 1 if a subject received KRAS testing at any point in the course of their clinical care, and 0 if there was no evidence of KRAS testing. The following characteristics were included in the model as binary variables: whether or not the subject: had surgery, received a referral to medical oncology, received an EGFR inhibitor, had a Medicare insurance product, and hada family history of cancer. The following characteristics were included in the model as continuous variables: age, BMI, survival (in years), average household income, average percentage of people with high school education or better in census tract, and year of metastatic diagnosis. Categorical variables included: number of chemotherapy regimens (categorized as 0, 1, 2, 3, 4 or 5+), Charlson comorbidity index (categorized as 0, 1, 2, 3, 4, or 5+) (11), cancer stage (categorized as III or IV), race (categorized as White, Asian, Black, Pacific Islander, American Indian, Other or Unknown), ethnicity (categorized as Hispanic or not Hispanic), gender (categorized as male or female), and integrated health system (categorized as A-G representing the 7 systems).

We used SAS (v9.2, SAS Institute, Cary NC) to perform all analyses. We produced descriptive statistics using PROC UNIVARIATE, MEANS and FREQ. We performed Wilcoxon Rank Sum tests using PROC NPAR1WAY. We used logistic regression (using PROC LOGISTIC) to model the likelihood of receiving KRAS testing. Using both forward and backward selection methods resulted in the same set of features being included in the model.

Results

A total of 4,446 patients met our eligibility criteria. To obtain more comprehensive data, a subset were selected for manual chart review, which was completed on 2,099 (47%) patient charts, including 1,152 (43% of 2,704) stage III CRC cases and 947 (54% of 1,742) stage IV CRC cases. We confirmed a diagnosis of mCRC for 1,188 patients (266 diagnosed at stage III, and 922 diagnosed at stage IV). This was the cohort used for the remaining analyses. With one exception, all sites conducted manual chart review on all (100%; 3 sites) or nearly all (over 90%; 3 sites) of the eligible cases. The remaining site, the largest, randomly selected a subset of 22% of charts for review due to resource constraints.

We identified a clinical KRAS test result for 428 patients with mCRC (36%) (Table 1). Of those tested, 40% had a mutation in codon 12 or 13 (Table 2), consistent with reported mutation frequencies from other studies (13, 14). There was insufficient biological tissue available for 465 patients (39%).

Table 1.

Treatment status, health status, and demographic features of subject who did and did not receive KRAS testing as part of their clinical care for colorectal cancer. P-values are unadjusted, from chi-square tests.

Characteristic	Value	Tested	Not Tested	p-value	Characteristic	Value	Tested	Not Tested	p-value
Chemotherapy Regimens	0	13	171	<.0001	Race	White	306	520	<.0001
	1	28	146			Asian	42	52
	2	75	99			Black	37	90
	3	117	53			Pacific Is.	13	11
	4	80	49			Am. Indian	2	4
	>5	264	93			Other	10	31
						Unknown	15	55
Surgery1	No	105	351	<.0001	Ethnicity	Not Hispanic	319	592	0.0984
	Yes	322	410			Hispanic	27	45
Referral to Med. Oncology	No	5	166	<.0001	Gender	Male	222	379	0.1368
	Yes	407	610			Female	205	381
BMI Category	<20	17	61	<.0001	Household Income	< $40K	95	221	0.0024
	20-24	89	165			$40K - $59K	142	263
	25-29	122	182			$60K - $79K	90	141
	≥ 30	114	150			≥ $80K	63	107
Age at diagnosis, years	< 50	63	75	<.0001	% High School Education	< 50%	4	12	<.0001
	50-59	124	110			50-69%	36	63
	60-69	143	149			70-89%	197	383
	70-79	73	230			≥ 90%	166	260
	≥ 80	22	199
Charlson co-morbidity Index	0	291	356	<.0001	Year of Metastatic Diagnosis	2006	50	211	<.0001
	1	170	176			2007	123	228
	2	54	80			2008	179	194
	3	12	18			2009	93	110
	4	8	13		Medicare	No	274	581	<.0001
	≥ 5	2	8			Yes	150	183
Any EGFR inhibitor	No	262	674	<.0001	Organization	A	44	57	<.0001
	Yes	164	88			B	170	256
Survival Interval, years (if deceased)	0.5	28	438	0.0010		C	57	92
	0.5–1	36	94			D	66	127
	1 – 2	73	131			E	37	87
	≥ 2	241	147			F	18	44
Family History of Cancer	No	129	220	<.0001		G	37	96
	Yes	214	296		Cancer Stage	III	120	146	<.0001
	Unk	86	243			IV	313	609

¹Colectomy or hemicolectomy

Table 2.

Association of KRAS testing and treatment with EGFR inhibitors among patients with mCRC.

	Treated with EGFR Inhibitors1	Not Treated with EGFR Inhibitors	Total
No KRAS Test	89 (35%)	671 (72%)	760 (64%)
KRAS Test2	163 (65%)	265 (28%)	428 (36%)
WT	136 (83%)	110 (42%)	246 (57%)
Mutation	25 (15%)	152 (57%)	177 (42%)3
Insufficient Sample	2 (1%)	3 (1%)	5 (1%)

¹Patient was treated with cetuximab or panitumumab at any time during the course of their clinical care.

²A KRAS test was ordered as part of the patient's clinical care. WT = wild-type genotype, Mutation = any KRAS mutation for any codon tested, primarily codons 12 and 13 for most patients. Insufficient sample = a test was ordered but no KRAS genotype was reported.

³40% had mutations in codon 12 or 13. An additional 2% had a mutation detected in codon 61. However, not all labs tested codon 61 as part of their KRAS sequencing protocol.

Trends in KRAS Test Utilization

Beginning in June, 2008, there was a striking increase in utilization of KRAS testing (Figure 1). The decline in the number of tests ordered toward the end of 2010 may be a product of study enrollment ending in December, 2009.

Figure 1.

Overview of KRAS Test Utilization for study participants.

Of those who received KRAS testing, the median time between mCRC diagnosis and KRAS testing was about 10 months, with a range of 0 days to 4 years. There were significant changes in this interval over time (Wilcoxon Rank Sum p-value < 0.0001). From 2006 to 2009, the median interval between mCRC diagnosis and KRAS testing declined from 2.2 years to 2 months. Over the same time frame, EGFR inhibitors were used earlier within the patient's course of clinical care, with a median interval from metastatic diagnosis date to treatment with an EGFR inhibitor (among those receiving EGFR inhibitors) of 25 months in 2006 and 7 months in 2009. The interval between KRAS testing and initiation of EGFR inhibitor treatment increased from one month in 2006 to about 3 months in 2009. The percentage of patients tested within 90 days of mCRC diagnosis (an arbitrary threshold) is significantly associated with year of mCRC diagnosis, rising from 5% in 2006 to 29% in 2009 (p < 0.0001) (Table 3).

Table 3.

Timing of KRAS test within the course of mCRC diagnosis and treatment among patients who received KRAS testing.

	N (%) Tested
Year of mCRC diagnosis	Before mCRC diagnosis	At mCRC diagnosis (<90 days)	Between diagnosis and treatment1	Before treatment1 (<90 days)	After treatment1 initiated	Total
2006	2 (6%)	3 (9%)	3 (9%)	13 (38%)	15 (44%)	34
2007	3 (5%)	4 (7%)	14 (24%)	16 (27%)	23 (39%)	59
2008	3 (5%)	19 (31%)	13 (21%)	21 (34%)	12 (19%)	62
2009	3 (12%)	13 (50%)	4 (15%)	9 (35%)	3 (12%)	26

¹treatment with EGFR inhibitors. Includes those tested but not yet treated.

^*KRAS test dates are unknown for 71 participants

KRAS Test Association with Patient Treatment

Overall, 252 patients (21%) received an EGFR inhibitor during the course of their clinical care. When looking solely at EGFR inhibitor use, the majority received cetuximab alone (214; 85%). The remainder received both cetuximab and panitumumab in different lines of therapy (26; 10%) or panitumumab alone (12; 5%). These drugs were received either alone or in combination with other therapies.

Most patients who received EGFR inhibitors (65%) also received KRAS testing, and nearly all patients who received EGFR inhibitors (83%) had a KRAS wild-type genotype. Of the 25 patients who had a KRAS mutation and yet received EGFR inhibitors, 16 were tested after EGFR inhibitor treatment was initiated, and an additional 7 were tested prior to mid-2008. Only two were tested and treated with an EGFR inhibitor after the recommendations went into place. Among those who received EGFR inhibitors, the year EGFR therapy was initiated was strongly associated with receipt of KRAS testing at any time, and also strongly associated with receipt of KRAS testing prior to initiating treatment with EGFR inhibitors (Figure 2; Table 3). By 2009, more than 96% of patients who received EGFR inhibitors were tested for KRAS status prior to initiating therapy.

Figure 2.

Percentage of patients who received KRAS testing before EGFR Inhibitor treatment (white bars) and at any time in their clinical care (black bars) by year that EGFR inhibitor treatment was initiated, among patients who received EGFR inhibitors.

In contrast, among those who did not receive treatment with EGFR inhibitors (n=936), only 265 (28%) were tested for KRAS. There were 152 (57%; out of 265) patients with a KRAS mutation who did not receive EGFR inhibitors, consistent with current recommendations. There were 110 (42%) patients with KRAS wild-type genotype who did not receive EGFR inhibitors. These KRAS wild-type patients were diagnosed more recently (median mCRC diagnosis month of August 2008 versus December 2007; Wilcoxon Rank Sum p-value=0.0026), had received KRAS testing more recently (median KRAS test month December 2009 versus May 2009; p=0.0036), and had received fewer lines of therapy (mean number of 1.7 versus 2.4 lines of therapy; Wilcoxon Rank Sum p-value < 0.0001) compared with patients with KRAS wild-type genotype who received EGFR inhibitors. These patients had similar age at diagnosis (p=0.71), BMI (p=0.82), comorbidity index (p=0.58), interval to KRAS testing (p=0.38), gender (p=0.35), race (p=0.90), ethnicity (p=0.35), Medicare status (p=0.91), and length of survival (p=0.07) compared with patients with KRAS wild-type genotype who received EGFR inhibitors. Patients with KRAS wild-type who had not yet received EGFR inhibitors are at an earlier point in their cancer treatment, and may have received EGFR inhibitors after the close of the study window or have yet to receive them.

Most patients (60%) received EGFR inhibitors as part of their last reported line of therapy; 22% received EGFR inhibitors as a middle line treatment, and 18% as their first line treatment. For patients diagnosed prior to mid-2008, only 45%, 53%, and 63% of patients who received EGFR inhibitors as their first line, middle line, or last line of therapy, respectively, ever received KRAS testing. For patients diagnosed after mid-2008, the proportion who received KRAS testing increased to 82%, 88%, and 94% of patients who received EGFR inhibitors as their first line, middle line, or last line of therapy, respectively.

Patient Characteristics

Several aspects of patients' treatment history, health status, and demographics were significantly associated with KRAS testing status.

Treatment History

Patient treatment history was a strong predictor of receipt of KRAS testing (Table 4). Patients who did not receive any chemotherapy were significantly less likely to receive KRAS testing than patients who received at least one line of chemotherapy (7% versus 60%, adjusted p < 0.0001). Patients receiving more lines of chemotherapy were more likely to receive KRAS testing. The remaining treatment history variables that we considered were no longer significant after adjusting for cancer stage, number of lines of therapy, age at diagnosis, survival interval, Charlson comorbidity index, and metastatic diagnosis date.

Table 4.

Multivariate modeling of relationship of mCRC patient characteristics to receipt of KRAS testing as part of clinical care.

Treatment Status Characteristics						Demographic Characteristics
Characteristic	Value	N	% Tested	OR (95% CI)	p-value2	Characteristic	Value	N	% Tested	OR (95% CI)	p-value
Chemotherapy Regimens	0	184	7	0.02 (0.003, 0.10)	<.0001	Race	White	826	37	--	0.8221
	1	174	16	0.08 (0.06, 0.10)			Asian	94	45	0.6 (0.3, 1.2)
	2	174	43	0.43 (0.33, 0.57)			Black	127	29	0.8 (0.4, 1.7)
	3	170	69	1.05 (0.86, 1.30)			Pacific Is.	24	54	1.0 (0.3, 3.0)
	4	129	62	0.41 (0.36, 0.44)			Am. Indian	6	33	0.8 (0.1, 7.9)
	≥5	357	74	--			Other	41	24	1.6 (0.4, 2.8)
							Unknown	70	21	1.0 (0.4, 2.5)
Surgery1	No	456	23	0.63 (0.40, 0.91)	0.0187	Ethnicity	Not Hispanic	911	35	--	0.5002
	Yes	732	44	--			Hispanic	72	38	0.7 (0.3, 1.4)
Referral to Med. Oncology	No	171	3	0.77 (0.21, 2.5)	0.6569	Gender	Male	601	37	--	0.5899
	Yes	1017	40	--			Female	586	35	0.8 (0.6, 1.2)
Any EGFR inhibitor	No	936	28	--	0.0014	Household Income	< $40K	316	30	0.8 (0.5, 1.4)	0.4753
	Yes	252	65	2.0 (1.3, 3.1)			$40K - $59K	405	35	--
Health Status Characteristics							$60K - $79K	231	39	1.3 (1.1, 1.4)
Characteristic	Value	N	% Tested	OR (95% CI)	p-value		≥ $80K	170	37	1.2 (1.1, 1.5)
Age at diagnosis, years	< 50	138	46	1.4 (0.8, 3.0)	0.0034	% High School Education	< 50%	16	25	0.7 (0.1, 3.3)	0.9924
	50-59	234	53	2.4 (2.3, 3.0)			50-69%	99	36	1.4 (1.0, 1.8)
	60-69	292	49	1.7 (1.3, 2.5)			70-89%	580	34	--
	70-79	303	24	--			≥ 90%	426	39	1.2 (1.0, 1.4)
	≥ 80	221	10	0.4 (0.3, 0.7)
Charlson Co-morbidity Index	0	647	45	--	0.0316	Year of Metastatic Diagnosis	2006	261	19	0.06 (0.02, 0.1)	<.0001
	1	346	49	0.5 (0.1, 1.4)			2007	351	35	0.2 (0.2, 0.5)
	2	134	40	0.3 (0.1, 0.8)			2008	373	48	--
	3	30	41	0.5 (0.2, 1.7)			2009	203	46	0.9 (0.7, 1.1)
	4	21	38	0.6 (0.2, 2.0)		Medicare	No	855	32	--	0.1883
	≥ 5	10	20	0.3 (0.1, 0.8)			Yes	333	45	1.8 (0.9, 2.2)
BMI Category	<20	78	22	0.9 (0.4, 2.0)	0.7945	Organization	A	101	44	0.7 (0.4, 1.1)	0.0220
	20-24	254	35	0.9 (0.8, 1.0)			B	426	40	--
	25-29	304	40	--			C	149	38	0.6 (0.5, 0.8)
	≥ 30	264	43	1.2 (1.1, 1.3)			D	193	34	0.6 (0.4, 0.8)
Survival Interval, years (if deceased)	0.5	466	6	--	0.0010		E	124	30	0.3 (0.2, 0.4)
	0.5–1	130	28	1.9 (0.8, 4.3)			F	62	29	0.6 (0.4, 0.8)
	1 – 2	204	36	2.9 (1.4, 6.3)			G	133	28	1.0 (0.6, 1.3)
	≥ 2	388	62	5.1 (2.3, 11.4)		Family History of Cancer	No	349	37	0.9 (0.6, 1.3)	0.5208
Cancer Stage	III	266	45	1.7 (1.1, 2.5)	0.0078		Yes	510	42	--
	IV	922	34	--			Unknown	329	26	0.7 (0.4, 0.9)

OR=odds ratio. CI= confidence interval. The largest group in each category is used as the basis of the OR.

¹Colectomy or hemicolectomy

²All reported Wald chi-square p-values and ORs are from a logistic regression model including all factors in Table 4, and a binary variable indicating presence or absence of KRAS testing as the response variable. Since the reported percent tested is based on the raw, unadjusted counts, there can be apparent discrepancies between the percent tested and the reported adjusted ORs.

Health Status

Patient overall health status was another significant determinant of receipt of KRAS testing (Table 4). Patients with increasing age (p = 0.003), more co-morbidities (p = 0.03), or who were deceased within 6 months of diagnosis with mCRC (p = 0.001) were significantly less likely to receive KRAS testing after adjustment for other factors. These patients were also significantly less likely to receive any form of chemotherapy (e.g., 33% versus 76% received any chemotherapy for those over/under 80 years old, respectively; 50% versus 76% received any chemotherapy for those with more/less than 5 co-morbidities, respectively; 61% versus 72% received any chemotherapy for those with less/more than 6 months survival following diagnosis with mCRC, respectively).

Demographics

None of the demographic characteristics were associated with receipt of KRAS testing after adjusting for treatment history, clinical characteristics, and health status (Table 4).

Provider Characteristics

For three sites, we were able to collect information on the ordering physician. Across these sites, 89% (117 out of 131) of medical oncologists ordered at least one KRAS test. Because the KRAS test is so widely adopted among clinicians at these sites, we did not explore provider characteristics further, since this is unlikely to be a major explanation for practice variation in this study.

System Characteristics

Insurance Product

Although there is limited variation on type of insurance for patients in this study, patients with Medicare coverage were more likely to receive KRAS testing than patients without this coverage; however, that difference appears to be explained by other clinical factors, as evidenced by the insignificant Wald chi-square p-value in the multivariate modeling (45% versus 32%, adjusted p = 0.19).

Geographic Distribution

We observed significant variation across sites in the proportion of patients who received KRAS testing ranging from 28% to 44% (p = 0.02) (Figure 3). There was more substantial variation in the median interval between diagnosis and receipt of KRAS testing by site ranging between about three months and one year (Wilcoxon rank sum test p-value <0.0001).

Figure 3.

System characteristics: Practice variation in (A) the proportion of patients who received KRAS testing and (B) the median test interval (number of days between diagnosis with mCRC and receipt of KRAS testing) across integrated health systems participating in this study.

Discussion

This study illustrates the remarkably rapid and wide adoption of KRAS testing by clinicians in integrated health care delivery settings. In these settings, KRAS testing programs began in mid-2008 prior to changes in professional or regulatory guidance. Since 2009, over 94% of patients who received EGFR inhibitors were tested for KRAS status prior to initiating therapy. Eighty-three percent of patients treated with EGFR inhibitors had a KRAS wild-type genotype. Most patients with a KRAS mutation (86%) were not treated with EGFR inhibitors. Almost 90% of oncologists in our study ordered at least one KRAS test. Although overall only 36% of the patients in the study population received KRAS testing, these decisions appear to be personalized and tailored to individual patient characteristics. Major factors influencing testing were related to patient treatment history – e.g., patients who did not receive any chemotherapy were less likely to be tested – and also to overall health status – those who were elderly, had multiple co-morbidities, or who were deceased within 6 months of mCRC diagnosis were less likely to be tested.

Two events immediately preceded the rapid increase in KRAS testing beginning in June 2008. First, at ASCO's annual meeting in late May 2008, retrospective analyses of the CRYSTAL (15) and OPUS (16) randomized clinical trials were presented demonstrating the lack of effect for EGFR inhibitors in patients with KRAS mutations. Second, the European Medicines Agency (EMEA) Committee for medicinal products for human use (CHMP) issued an opinion in May 2008 indicating the use of EGFR inhibitors for treatment of mCRC patients who are KRAS wild-type (17). It is remarkable that this increase occurred prior to release of guidance in the U.S. from ASCO, the FDA, or the NCCN.

Our findings illustrate a low risk for disparities in access to testing and/or treatment in these settings. Although measures of socio-economic status (i.e., household income, education, and Medicare status) were associated with KRAS testing status in unadjusted comparisons, these associations were no longer significant after adjustment for other patient characteristics. We did observe unexplained variation across sites in the proportion of patients who receive KRAS testing and in the timing of KRAS testing, suggesting that system-wide factors (e.g., institutional policies, treating physician preference) may influence test use even after accounting for the patient's disease status and treatment.

The variation across sites has practical implications for the effectiveness of the KRAS testing program. Ideally, the KRAS test will only be used for patients who are eligible and considered for this treatment, but most patients (up to 80% in this study) may never receive EGFR inhibitors. Since the interval between mCRC diagnosis and KRAS testing is decreasing over time, we evaluated whether the sites with the shorter median testing interval were simply late adopters of KRAS testing. Although this may be an explanation for one site with the shortest median interval, it does not explain the overall trend across the remaining sites.

The primary limitation of this study is that it was conducted only within integrated health care delivery systems, which may differ significantly from other types of delivery systems. An important future research goal is to evaluate utilization of KRAS testing in other delivery settings in the US. Outside of the US, a recent report evaluated the uptake of the KRAS test in Asia, Latin America, and Europe (18). They also noted rapid uptake of testing from 3% in 2008 to 69% in 2010, with significant regional variation of 78%, 63%, and 44% receiving KRAS testing in 2010 in Europe, Latin America, and Asia, respectively. The majority of patients in the Ciardiello study were selected for inclusion because they were receiving chemotherapy. Our results are not precisely comparable because we included all or nearly all (>90%) patients with mCRC regardless of treatment status, and 32% of patients in our study did not receive any chemotherapy. Of the people who received any chemotherapy in 2010, 78% were tested for KRAS status.

Other limitations are related to the retrospective, observational nature of the study design. We did not collect information directly from patients and thus our findings are based solely on information from medical records and KRAS testing vendors. There is variation in the available follow-up time since patients diagnosed in 2009 were more likely to be censored. Also, there may be other unmeasured confounders that impact the interpretation of study results. For example, complications or poor health status not fully captured by our comorbidity measure also may have contraindicated chemotherapy.

The environment was also a strength of our study design. We leveraged the CRN research infrastructure to draw upon our coordinated systems of electronic data and tumor registries. Consistency of file structures and data definitions allows us to have efficiency in multi-institution research, such as facilitating use of the same criteria across sites in case definition and selection. Our work also highlights limitations to our research databases and electronic medical records. Importantly, 1) we had to use manual chart review to identify progression to mCRC (these study data will be used to develop and validate better prediction algorithms for future work), and 2) we were not able to identify either who received the genetic test or the test results from electronic sources. Improved coding and recording of genetic test orders and results would greatly facilitatefuture studies.

The availability of new technologies is not usually sufficient for their optimal implementation in clinical practice (19-21). For example, utilization rates are typically low (around 30%) for testing of inherited mutations in the BRCA1/2 gene even among patients at high risk for breast cancer who are referred to genetic counseling (22-25). Observational studies such as ours have a larger sample size, a greater diversity of patients, and potentially a greater diversity of care models compared to the clinical trial settings that formed the basis of the professional recommendations. Thus, this study informs the ultimate purpose of improving patient outcomes and reducing population health disparities.

This study was one part of the CERGEN project, a comprehensive research program in comparative effectiveness research for genomic applications. This includes interviews with patients, providers, and health plan decision makers regarding the implementation of KRAS testing. This methodology may allow us to explore some of the unexplained practice variation across sites that we observed in this study. We also plan to compare outcomes for patients whose treatment decisions are guided by KRAS testing versus patients who did not receive testing. Through comparative effectiveness research, we have been able to describe the sequence of diffusion of mCRC treatment decision making guided by KRAS testing, made more valuable by the multi-site and wide diversity of patients and geographic diversity of providers.