The role of health, sociodemographic, and care delivery factors in timely completion of colonoscopy in a US-based primary care population: a retrospective analysis

Abstract

Background

Timely colonoscopy completion in primary care can prevent diagnostic delays in colorectal cancer. Factors that influence why patients experience timely or delayed colonoscopy completion are unclear. We sought to identify potentially intervenable factors associated with earlier (or later) colonoscopy test completion in primary care.

Methods

All colonoscopy orders placed by primary care clinicians in two clinics within a single hospital system between January 1 2018 and December 31 2021 were examined for time to completion using the Cox Proportional Hazards Model, where the hazards of completion were adjusted for variables potentially associated with the outcome, including sociodemographic, individual health-related, and care delivery factors.

Results

Among 10,576 colonoscopy tests ordered, 56% were completed within one year and the median time to colonoscopy completion was 230 days, 95% CI [217, 242]. After multivariable adjustment, earlier colonoscopy completion was associated with receiving care at a community health center (HR 1.13, 95% CI [1.03, 1.24]), preferred spoken language other than English (HR 1.23, 95% CI [1.12, 1.34]), male sex (HR 1.09, 95% CI [1.03, 1.14]), Black race (HR 1.08, 95% CI [1.02, 1.16]), any college education (HR 1.07, 95% CI [1.01, 1.14]), a diagnosis of rectal bleeding (HR 1.88, 95% CI [1.69, 2.08]), and documented use of an electronic patient portal (HR 1.19, 95% CI [1.13, 1.26]). Completion occurred later among patients with Medicaid insurance (HR 0.82, 95% CI [0.75, 0.89]), subsidized commercial insurance (HR 0.80, 95% CI [0.69, 0.91]), depression (HR 0.95, 95% CI [0.89, 1.01]), and when ordered by a nurse practitioner (HR 0.77, 95% CI [0.68, 0.88]), resident (HR 0.92, 95% CI [0.85, 0.99]) or during a telehealth appointment (via telephone HR 0.56, 95% CI [0.49, 0.64] and via video HR 0.76, 95% CI [0.63, 0.90]). Results were similar for a sensitivity analysis restricted to patients within a Medicaid Accountable Care Organization (ACO) network.

Conclusions

Several sociodemographic, clinical, and telehealth factors were associated with time to colonoscopy completion, identifying potential opportunities for targeted care and future research.

Background

Differences in colorectal cancer (CRC) outcomes according to population characteristics are well documented [1–4], although the mechanisms remain unclear [5]. Delayed or missing CRC screening, delayed diagnostic colonoscopy after abnormal fecal immunohistochemistry testing (FIT), and delayed treatment have been repeatedly associated with worse CRC outcomes [6–9]. Limited efforts have been made to understand drivers of delayed colonoscopy completion in primary care, and prior studies have primarily explored barriers to completing diagnostic testing [10, 11]. As part of a multi-year study focused on improving test loop closure in primary care, we sought to understand the factors associated with earlier or later completion of ordered colonoscopies (screening and diagnostic) among primary care patients within a single hospital system in the United States (US). We explored whether certain sociodemographic, health, and care delivery factors were positively or negatively associated with time to colonoscopy completion, to identify factors that could be impacted by improved primary care processes. We suggest future pragmatic interventions and investigations to improve timely completion of colorectal cancer tests within US-based primary care.

Methods

To assess factors associated with timeliness of colonoscopy completion, we included patients who had colonoscopy orders placed by primary care clinicians at two practices within in the same US-based hospital system between January 1, 2018, and December 31, 2021. The two practices included a Community Health Center (CHC) practice –with enhanced patient and care navigation supports- and a hospital-based academic practice. Colonoscopy tests ordered less than one year prior to the date of last follow up (February 12, 2022) were excluded from analysis to ensure sufficient observation time. Demographic, health-related, and care delivery data were derived from electronic health record information that was stored in a central data warehouse for the hospital system.

In the primary analysis (n = 10,576 tests ordered), we used Cox proportional hazards models to assess factors associated with time to colonoscopy completion. Incomplete colonoscopy orders were administratively censored at the end of the period of observation (February 12, 2022). Models were adjusted for sociodemographic (insurance type, highest education level, ethnicity, language, race), health-related (age, sex, depression diagnosis, health complexity, rectal bleeding diagnosis, electronic patient portal use), and care delivery factors (practice site, visit modality, ordering clinician, and ordered during the COVID pandemic -defined as March 1, 2020 through December 31, 2021). We assessed the proportional hazards assumption using smoothed Schoenfeld residual plots. In addition to the primary analysis, we performed a sensitivity analysis in a subgroup of Medicaid Accountable Care Organization (ACO) patients (n = 1257 tests ordered) whose insurance requires tests to be completed within the hospital system represented in our EMR-derived dataset. Repeating the analysis in this subgroup allows us to control for the effect of unmeasured confounders in the insurance type category and to exclude the possibility of colonoscopy completion outside the hospital system for patients with non-ACO insurance products. We report hazard ratios (HRs) with 95% confidence intervals (CIs).

Results

Among 10,576 colonoscopy tests ordered from 2018 to 2021, 44% of colonoscopies were not completed within a year after ordering. The median time to colonoscopy completion was 230 days, 95% CI [217, 242]. Table 1 summarizes population characteristics. Table 2 presents adjusted hazard ratios (HRs) from the Cox models. Across all analyses, smoothed Schoenfeld residual plots showed no evidence of violation of the proportional hazards assumption. (Table 1) (Table 2).

Table 1.

Baseline characteristics of the study population

Factor	Category	Main Analysis Population N (%)	Sensitivity Analysis Population N (%)
Total Population		10,576	1,257
Age (Mean, SD)a		59.39 (10.47)	55.94 (8.89)
Sex	Female	5,565 (52.6)	754 (60.0)
	Male	5,011 (47.4)	503 (40.0)
English Language	Yes	9,159 (86.6)	791 (62.9)
	No	1,378 (13.0)	457 (36.4)
	Missingb	39 (0.4)	9 (0.7)
Race	White	5,885 (55.6)	358 (28.5)
	Asian	524 (5.0)	76 (6.0)
	Black	3,081 (29.1)	574 (45.7)
	Other	672 (6.4)	172 (13.7)
	Missingb	414 (3.9)	77 (6.1)
Hispanic	No	9,066 (85.7)	1,015 (80.7)
	Yes	915 (8.7)	174 (13.8)
	Missingb	595 (5.6)	68 (5.4)
Insurancek	Commercial	5,670 (53.6)	-
	Medicaid only	1,257 (11.9)	-
	Medicare only	2,750 (26.0)	-
	Medicare and Medicaid	358 (3.4)	-
	Other	160 (1.5)	-
	Subsidized commercial	381 (3.6)	-
Educationc	High school or less	4,328 (40.9)	790 (62.8)
	Greater than high school	4,946 (46.8)	297 (23.6)
	Missingb	1,302 (12.3)	170 (13.5)
Depressiond	No	8,559 (80.9)	932 (74.1)
	Yes	2,017 (19.1)	325 (25.9)
Charlson Comorbidity Scoree	0	5,652 (53.4)	637 (50.7)
	1–2	3,038 (28.7)	367 (29.2)
	3–4	971 (9.2)	129 (10.3)
	5 or higher	915 (8.7)	124 (9.9)
Colonoscopy Ordered Byf	Attending	8,542 (80.8)	878 (69.8)
	Nurse Practitioner	448 (4.2)	58 (4.6)
	Resident	1,586 (15.0)	321 (25.5)
Portal Useg	No	7,084 (67.0)	1,091 (86.8)
	Yes	3,492 (33.0)	166 (13.2)
Colonoscopy Ordered During COVIDh	No	7,322 (69.2)	878 (69.8)
	Yes	3,254 (30.8)	379 (30.2)
Index Visit Modalityi	In person	8,158 (77.1)	1,010 (80.4)
	Telephone	705 (6.7)	100 (8.0)
	Video	293 (2.8)	12 (1.0)
	Missingb	1,420 (13.4)	135 (10.7)
Rectal Bleedingj	Yes	574 (5.4)	88 (7.0)
	No	9,991 (94.5)	1,167 (92.8)
	Missingb	11 (0.1)	2 (0.2)
Practice Site	Community Health Center	1,269 (12.0)	367 (29.2)
	Hospital-based practice	9,307 (88.0)	890 (70.8)
Colonoscopy Completion in 1 Yearl	No	4,628 (43.8)	624 (49.6)
	Yes	5,948 (56.2)	633 (50.4)

a. SD denotes standard deviation

b. Missing refers to the absence of data in the electronic health record

c. Patients without education data were coded as “missing” Those with a high school diploma or less were categorized as “high school or below”, and all others as “greater than high school.”

d. Depression was defined using ICD-10 codes (F32.xx-F33.xx)

e. Charlson Co-morbidity Index [25] was calculated from billing diagnoses for all encounters within 2 years of the index referral date. Patients were classified into 3 categories based on their score: 0, 1–2, 3–4 and 5 or higher

f. Colonoscopy orders were placed by attending physicians, nurse practitioners, and residents

g. Patient portal use was defined as registration on the patient site and viewing at least one note

h. We defined colonoscopy ordering during COVID as colonoscopy ordered between March 1st, 2020 and December 31st, 2021

i. Visit modality was separated into in-person, video, or telephone

j. Rectal bleeding was defined using ICD-10 codes (K62.5, K92.2, K51.911)

k. ‘Medicare’ includes all Medicare products (Advantage and traditional). ‘Medicaid’ includes all Masshealth products including Masshealth ACO, Masshealth Limited, and Masshealth Standard. ‘Commercial’ includes all HMO and PPO insurance plans excluding Medicare Advantage plans and subsidized “Connector” plans (i.e. ‘Subsidized’). ‘Medicare/Medicaid’ includes those with dual coverage. ‘Other’ includes all other plans not falling into above categories

l. Colonoscopy completion in 1 year was defined as completion within 365 days of ordering

Table 2.

Multivariable adjusted hazard ratios (HRs) for colonoscopy completion

Factor	Main Analysis HR (95% CI)	Sensitivity Analysis HR (95% CI)
Agea	1.00 [0.96, 1.05]	1.00 [0.99, 1.02]
Sex
Female (Reference)	1.00	1.00
Male	1.09 [1.03, 1.14]	1.00 [0.85, 1.17]
Hispanic
No (Reference)	1.00	1.00
Yes	1.01 [0.91, 1.12]	0.94 [0.74, 1.20]
Missingb	0.91 [0.80, 1.03]	0.53 [0.32, 0.87]
Education c
High school or less (Reference)	1.00	1.00
Greater than high school	1.07 [1.01, 1.14]	1.18 [0.97, 1.44]
Missingb	1.01 [0.93, 1.10]	0.97 [0.75, 1.24]
Depression d
No (Reference)	1.00	1.00
Yes	0.95 [0.89, 1.01]	0.92 [0.76, 1.10]
Charlson Comorbidity Score e
0 (Reference)	1.00	1.00
1–2	1.00 [0.94, 1.06]	0.95 [0.80, 1.13]
3–4	1.04 [0.95, 1.13]	0.88 [0.67, 1.15]
5 or higher	0.93 [0.84, 1.02]	0.77 [0.58, 1.03]
Colonoscopy Ordered By f
Attending (Reference)	1.00	1.00
Nurse Practitioner	0.77 [0.68, 0.88]	0.92 [0.63, 1.35]
Resident	0.92 [0.85, 0.99]	1.08 [0.89, 1.31]
Patient Portal Use g
No (Reference)	1.00	1.00
Yes	1.19 [1.13, 1.26]	1.03 [0.81, 1.32]
Colonoscopy Ordered During COVID h
No (Reference)	1.00	1.00
Yes	0.93 [0.87, 0.99]	1.08 [0.90, 1.31]
Index Visit Modality i
In person (Reference)	1.00	1.00
Telephone	0.56 [0.49, 0.64]	0.52 [0.35, 0.75]
Video	0.76 [0.63, 0.90]	0.66 [0.27, 1.63]
Missingb	1.15 [1.07, 1.24]	1.10 [0.85, 1.42]
Rectal Bleeding j
No (Reference)	1.00	1.00
Yes	1.88 [1.69, 2.08]	1.65 [1.24, 2.20]
Missing	1.46 [0.69, 3.07]	4.19 [1.03, 17.11]
Race
White (Reference)	1.00	1.00
Asian	1.02 [0.90, 1.14]	1.06 [0.76, 1.48]
Black	1.08 [1.02, 1.16]	0.98 [0.80, 1.20]
Other	1.03 [0.91, 1.16]	1.06 [0.81, 1.40]
Missingb	0.91 [0.78, 1.07]	0.95 [0.61, 1.50]
Insurance k
Commercial (Reference)	1.00	—
Medicaid	0.82 [0.75, 0.89]	—
Medicare	1.01 [0.94, 1.08]	—
Medicare and Medicaid	0.86 [0.74, 0.99]	—
Other	0.63 [0.50, 0.79]	—
Subsidized	0.80 [0.69, 0.91]	—
Practice Site
Hospital-based practice (Reference)	1.00	1.00
Community Health Center	1.13 [1.03, 1.24]	1.32 [1.07, 1.62]
English Language
Yes (Reference)	1.00	1.00
 No	1.23 [1.12, 1.34]	1.16 [0.95, 1.41]
Missingb	0.77 [0.46, 1.27]	1.03 [0.37, 2.85]

a. SD denotes standard deviation

b. Missing refers to the absence of data in the electronic health record

c. Patients without education data were coded as “missing” Those with a high school diploma or less were categorized as “high school or below”, and all others as “greater than high school.”

d. Depression was defined using ICD-10 codes (F32.xx-F33.xx)

e. Charlson Co-morbidity Index [25] was calculated from billing diagnoses for all encounters within 2 years of the index referral date. Patients were classified into 3 categories based on their score: 0, 1–2, 3–4 and 5 or higher

f. Colonoscopy orders were placed by attending physicians, nurse practitioners, and residents

g. Patient portal use was defined as registration on the patient site and viewing at least one note

h. We defined colonoscopy ordering during COVID as colonoscopy ordered between March 1st, 2020 and December 31st, 2021

i. Visit modality was separated into in-person, video, or telephone

j. Rectal bleeding was defined using ICD-10 codes (K62.5, K92.2, K51.911)

k. ‘Medicare’ includes all Medicare products (Advantage and traditional). ‘Medicaid’ includes all Masshealth products including Masshealth ACO, Masshealth Limited, and Masshealth Standard. ‘Commercial’ includes all HMO and PPO insurance plans excluding Medicare Advantage plans and subsidized “Connector” plans (i.e. ‘Subsidized’). ‘Medicare/Medicaid’ includes those with dual coverage. ‘Other’ includes all other plans not falling into above categories

l. Colonoscopy completion in 1 year was defined as completion within 365 days of ordering

In the primary analysis using a Cox proportional hazards model, earlier completion of colonoscopy was associated with several factors - higher education, non-English language preference, Black race, male sex, rectal bleeding diagnosis, electronic patient portal use, and care at the community health center practice. Patients with greater than high school education were more likely to complete colonoscopy earlier than those with high school or less (HR 1.07, 95% CI [1.01, 1.14]). Compared to female patients, male patients were more likely to complete colonoscopy earlier (HR 1.09, 95% CI [1.03, 1.14]). Non-English-speaking patients were also more likely to complete colonoscopy earlier than English speakers (HR 1.23, 95% CI [1.12, 1.34]). Black patients were more likely than White patients to complete colonoscopy earlier (HR 1.08, 95% CI [1.02, 1.16]). Patients with rectal bleeding and those using the patient portal had a substantially higher likelihood of earlier completion (HR 1.88, 95% CI [1.69, 2.08] and HR 1.19, 95% CI [1.13, 1.26], respectively). Similarly, patients receiving care at the community health center practice were significantly more likely to complete colonoscopy earlier than those seen in the hospital-based practice (HR 1.13, 95% CI [1.03, 1.24]), as shown in Fig. 1, which displays unadjusted cumulative probability of completing a colonoscopy test during the period of follow-up by practice setting (p < 0.01). (Figure 1)

Fig. 1.

Time to colonoscopy shown as cumulative probability, comparing the community health center and hospital-based practice

By contrast, orders placed by nurse practitioners (HR 0.77, 95% CI [0.68, 0.88]) or residents (HR 0.92, 95% CI [0.85, 0.99]) and those ordered at a telehealth visit (via telephone HR 0.56, 95% CI [0.49, 0.64] and via video HR 0.76, 95% CI [0.63, 0.90]) were associated with later completion compared to those placed by attending physicians and at in-person appointments, respectively. Compared to patients with commercial insurance, later test completion was observed for those with Medicaid (HR 0.82, 95% CI [0.75, 0.89]), dual coverage (Medicaid and Medicare) (HR 0.86, 95% CI [0.74, 0.99]), or subsidized commercial insurance (HR 0.80, 95% CI [0.69, 0.91]). Additionally, colonoscopies ordered during the COVID pandemic were more likely to be completed later than those ordered before the pandemic (HR 0.93, 95% CI [0.87, 0.99]).

The sensitivity analysis results (i.e. Medicaid ACO patients only) are consistent in directionality with those of the primary analysis, with the exception that Black patients no longer had higher likelihood of earlier test completion (HR 0.98, 95% CI [0.80, 1.20]).

Discussion

From 2018 to 2021 in this US-based hospital system, 56% of 10,576 colonoscopies ordered in primary care were completed within one year, indicating a need for primary care-based interventions to improve timely colonoscopy completion as has been discussed elsewhere [12, 13]. For patients in our cohort who completed their colonoscopy during the period of follow up, several unique factors that modify the rate of completion can inform future interventions and research. For instance, earlier colonoscopy completion among CHC patients -adjusted for race, language, ethnicity and education- compared to patients seen at a hospital-based practice is a novel finding. We hypothesize that this finding relates to organizational cultural competency at the CHC that historically has provided enhanced support, reminders, and instruction around colonoscopy scheduling and completion compared to the hospital-based primary care practice. In prior studies, patient navigation was an effective strategy to increase colonoscopy test completion in select high risk populations [14, 15]. Understanding and describing how CHCs intentionally support and help patients navigate the process of getting a diagnostic test can delineate best practices to guide other primary care sites. The slower completion of colonoscopy tests when ordered at phone and video appointments or by NPs or resident MDs –as our research group has shown elsewhere for a range of test orders [16]- should prompt practices to enhance test coordination efforts –assisted scheduling, reminder calls- after virtual visits or when ordered by a NP or resident. We also observed that patients with subsidized –high deductible- commercial insurance completed tests much later than other commercial insurance types. This finding is in keeping with studies that have shown barriers to CRC screening among uninsured or underinsured patients [17], indicating a need for research to explore nuances in the relationship between health disparities and insurance status.

Our focus on colonoscopy -screening and diagnostic- shapes and limits the implications of our findings. Although our analysis adjusted for rectal bleeding diagnosis to account for diagnostic colonoscopies ordered within our sample, we cannot otherwise reliably distinguish diagnostic from screening tests due to inconsistent diagnostic coding within our dataset. Relying on EHR documentation of test completion (vs. claims) in our sample could have missed tests completed outside the hospital system. To address both the risk of missing completed tests and of unmeasured confounding within insurance categories, we repeated the analysis within the Medicaid Accountable Care Organization (ACO) population separately. When we repeated the analysis for Medicaid ACO patients who are restricted to completing tests within the index hospital system, we observed consistent directionality with point estimates in the primary analysis, except for Black patients who no longer had higher likelihood of earlier test completion. The sensitivity analysis sample was substantially smaller (1,257 encounters vs. 10,576 in the main analysis) and included 574 (45.7%) Black patients. The loss of statistical significance and change in directionality among Black patients may reflect reduced statistical precision due to the smaller sample size or removal of unmeasured confounders that were associated with insurance status and race in the full analysis.

Other types of colon cancer tests -e.g. home-based screening tests such as fecal immunohistochemistry (FIT) or multitarget stool DNA (mt-sDNA) were not included in our dataset. Hence, the observation that non-English speakers are more likely to have earlier colonoscopy completion in our cohort may reflect factors influencing the likelihood of ordering a colonoscopy, as described elsewhere, including patient preference, provider bias, health literacy, availability of stool-based tests, and social support [18]. Because our study focuses on completion rates once ordered, we cannot exclude the possibility that clinicians may have offered and ordered colonoscopy (as opposed to other forms of colon cancer screening) for some patients in a systematically different way than for others–for example, based on the PCP’s assessment of their ability and likelihood to complete a colonoscopy. In that case, the denominator of non-English speakers with colonoscopy orders may have been systematically different from the denominator of all patients with a colonoscopy order. Comprehensive data regarding the rates of colon cancer screening across diverse patient populations, frequency of offering colon cancer screening along with shared decision making, and rates of “declined” screening are beyond the scope of this study, but are important contextual factors that may affect the likelihood of completion for all orders within our study.

Differences in primary care engagement and adherence to PCP recommendations between different ethnic groups –as have been shown elsewhere [19]- were not assessed in this study and may have contributed to more timely colonoscopy in the non-English speaking population in our sample. It may also be that trust between non-English speaking patients and their PCPs played a role in the recommendation of and likelihood to complete a colonoscopy. At the CHC practice, for instance, culturally concordant PCPs cared for a disproportionately higher percent of non-English speakers (average 60% among culturally concordant PCPs compared to 34%.) Studies of language concordant care generally show improved outcomes over language-discordant care [20]. Trust in PCPs has been shown to be a strong predictor of patient completion of colon cancer tests in other studies of racial and ethnic minorities and should be included in future analyses of colon cancer test completion [21]. Broadly, more study is needed to understand the interconnected processes of colon cancer screening test selection and test completion, particularly among non-English speaking populations.

More research is also needed to examine whether the positive relationship between patient portal use and earlier colonoscopy completion –as observed for other diagnostic testing loops [22]- is causal or a proxy for patient activation, as has been shown elsewhere [23, 24]. Increasing patient portal use may be a promising avenue for reducing the time to test completion. We are encouraged that a diagnosis of rectal bleeding is associated with earlier test completion, suggesting underlying system resiliency factors that may be instructive for improving colonoscopy test completion more generally.

We recognize that timely colonoscopy completion is one of several mechanisms that account for population differences in colorectal cancer health outcomes. Earlier test completion for some may not translate into earlier treatment or greater survival, due to underlying risks that are independent of screening or prompt diagnosis [5]. More research is needed to fill in the map of intervenable factors along the screening, diagnosis and treatment journey that optimize colorectal cancer outcomes.

Conclusions

The low rate (56%) of colonoscopy order completion within one year among primary care patients in a US-based hospital system underscores the need to identify, address, and harness factors that modify the timeliness of completion. Location of care (CHC practice), using an electronic patient portal, higher education, speaking a non-English language, and having a diagnosis of rectal bleeding were factors associated with earlier colonoscopy completion, whereas colonoscopy orders originating from a telehealth visit or ordered by a NP or resident, and in patients with depression, Medicaid, or subsidized commercial insurance, had later completion. Some of these factors lend themselves to systems interventions, and some stimulate novel hypotheses for research efforts to improve timely colonoscopy completion in diverse patient populations.

Abbreviations

ACO

Accountable Care Organization

CHC

Community Health Center

CRC

Colorectal cancer

EHR

Electronic health record

HR

Hazards ratio

NP

Nurse practitioner

US

United States

Authors’ contributions

TS, LF, RP, CD, and CZ jointly developed the research aim and methodology. CZ performed the Cox regression analyses and prepared Figure 1. TS drafted the manuscript. LF, AZ, RP, CD, and CZ provided key edits and suggestions.

Funding

Research was supported by AHRQ patient safety grant, R18HS027282.

Data availability

Datasets used for the analysis are available from the corresponding author upon reasonable request.

Declarations

Ethics approval and consent to participate

Per regulation 45 CFR 46.104(d) (2)(2), consent to participate was not required for this study of retrospective anonymized data where no human subjects were included. Our study adheres to the Declaration of Helsinki with regard to research conducted on human data. The data used for this research was approved by the BIDMC Committee on Clinical Investigation (2020P000502).

Consent for publication

Anonymized data was used in this study. Therefore, participant consent to publish was not necessary.

Competing interests

The authors declare no competing interests.