Currently Viewing:
The American Journal of Managed Care March 2013
Rates of Guideline Adherence Among US Community Oncologists Treating NSCLC
Zhaohui Wang, MD, PhD; Inga Aksamit, RN, MBA; Lisa Tuscher, BA; and Kim Bergstrom, PharmD
Effectiveness and Cost-Effectiveness of Diabetes Prevention Among Adherent Participants
William H. Herman, MD, MPH; Sharon L. Edelstein, ScM; Robert E. Ratner, MD; Maria G. Montez, RN, MSHP; Ronald T. Ackermann, MD, MPH; Trevor J. Orchard, MD; Mary A. Foulkes, PhD; Ping Zhang, PhD; Christopher D. Saudek, MD†; and Morton B. Brown, PhD; The Diabetes Prevention Program Research Group
Mental Health in ACOs: Missed Opportunities and Low-Hanging Fruit
Allison N. O'Donnell, MPH; Brent C. Williams, MD; Daniel Eisenberg, PhD; and Amy M. Kilbourne, PhD, MPH
Drug Adherence After Price Changes in a Previously Compliant Population
James J. Hill III, MD, MPH; Deron Galusha, MS; Martin D. Slade, MPH; and Mark R. Cullen, MD
Measuring Quality in the Early Years of Health Insurance Exchanges
Ledia M. Tabor, MPH; Phyllis Torda, MA; Sarah S. Thomas, MS; and Jennifer L. Zutz, MHSA
Currently Reading
Multilevel Predictors of Colorectal Cancer Screening Use in California
Salma Shariff-Marco, PhD, MPH; Nancy Breen, PhD; David G. Stinchcomb, MS, MA; and Carrie N. Klabunde, PhD
Factors Associated With Primary Hip Arthroplasty After Hip Fracture
Ishveen Chopra, MS; Khalid M. Kamal, PhD; Jayashri Sankaranarayanan, MPharm, PhD; and Gibbs Kanyongo, PhD
Measuring Concurrent Oral Hypoglycemic and Antidepressant Adherence and Clinical Outcomes
Hillary R. Bogner, MD, MSCE; Heather F. de Vries, MSPH; Alison J. O'Donnell, BA; and Knashawn H. Morales, ScD
Computed Tomography Scan Use Variation: Patient, Hospital, and Geographic Factors
Eric A. Vance, PhD; Xiaojin Xie, MS; Andrew Henry, BS; Christian Wernz, PhD; and Anthony D. Slonim, MD, DrPH
Low Clinical Utility of Folate Determinations in Primary Care Setting
Shlomo Vinker, MD; Eli Krantman, MD; Michal Shani, MD; and Sasson Nakar, MD
Trends in Inpatient Hospital Prices, 2008 to 2010
Jeff Lemieux, MA; and Teresa Mulligan, MHSA

Multilevel Predictors of Colorectal Cancer Screening Use in California

Salma Shariff-Marco, PhD, MPH; Nancy Breen, PhD; David G. Stinchcomb, MS, MA; and Carrie N. Klabunde, PhD
We studied contextual factors and found that locality, availability of primary care, and HMO membership influenced use of colorectal cancer screening in California.
To evaluate the effects of individual characteristics as well as the effects of area-level segregation, SES, and healthcare resources at the county and MSSA levels on individuallevel CRC screening use (by any modality and by FSIG), we constructed a series of multilevel logistic models using the gllamm function in Stata version 9 (StataCorp LP, College Station, Texas). As MSSAs are groups of census tracts and do not span county boundaries, we built models of individuals nested within MSSAs, which were nested within counties. We first developed an unconditional or “empty” model, which partitioned the variation in CRC screening use by each level (ie, individual, MSSA, and county), to assess the influence of contextual factors at different geographic scales on CRC screening use. We then added individual-level variables to assess how much of the variation could be explained by individual-level factors. Finally, we added MSSA and county characteristics to assess the effects of specific MSSA and county factors while controlling for the individual-level determinants of CRC screening use.

All variables were binary, consistent with the recommendation of Rabe-Hesketh and Skrondal.46 Because the CHIS sampling design oversamples in some counties, individual CHIS sample weights were first normalized and then scaled for the MSSA and county levels using the probabilityweighted iteratively generalized least squares weight-scaling method.47,48 The resulting scaled weights were then included in the multilevel model.

We present the area-level random effects (the between-area variation) using the measures previously suggested in the literature.49-52 The interclass correlation provided the between-area variance. The median odds ratio (MOR) adjusted the between-area variance to use the same scale as the individual-level odds ratio estimates. The MOR can be interpreted as the increased risk associated with an individual moving to an area with a higher risk. If the MOR was close to 1.0, there was little variation between areas; larger MOR values (2.0 or higher) indicated greater between-area variation. We report the proportional change in variance (PCV) for models 2 and 3 to show how much of the residual arealevel variance from the previous models was explained by the additional explanatory variables. The interclass correlation, MOR, PCV, and interval odds ratios were derived from model output following methods described by Larsen and colleagues (2000 and 2005).50,51 For ease of interpretation, we focus on MORs and PCVs in the Results section, but include the full set of output in the table of model results. For the area-level fixed effects (the within-area contextual effects), we report the interval odds ratio because it offers information on the magnitude of the fixed estimates in relation to its importance in explaining the residual area variance.

RESULTS

Characteristics of the Study Sample and Unadjusted Odds of CRC Screening

Table 1 presents the study sample’s characteristics. Column 3 shows that 60% of adults aged 50 to 84 years reported being up-to-date with any CRC screening modality. Among our sample, 38% were current for CRC screening by colonoscopy, 13% by FSIG, and 10% by FOBT. The majority of respondents were non-Hispanic white, between 50 and 64 years of age, married/living with a partner, insured, born in the United States, only spoke English at home, felt safe in their neighborhood all or most of the time, and reported no experiences of discrimination. They also had less than a bachelor’s degree, income at least 300% of the federal poverty level, and food security. Most lived in MSSAs and counties with greaterthan-the-median segregation levels, high HMO penetration, and without primary care shortages, and where less than 20% of the residents were living in poverty.

Columns 4 and 5 of Table 1 show unadjusted odds for being up-to-date with CRC screening by any modality and by FSIG. Respondents were at increased odds of being up-to-date with CRC screening by any modality and by FSIG if they were older, non-Hispanic white/African American, married, more educated, of higher income, insured, born in the United States, only spoke English at home, lived in a safe neighborhood, and had food security. For any modality, those who were male or reported no experiences of discrimination were at increased odds of being up-to-date. Regardless of modality, at both geographic levels respondents living in areas with higher poverty rates or primary care shortages had decreased odds of being up-to-date with CRC screening. For FSIG, less HMO penetration was associated with lower odds of being up-to-date. Higher MSSA segregation was associated with lower odds of being up-to-date by any modality.

The Figure shows the geographic variation in CRC screening use by modality and geographic levels. Median unadjusted rates of CRC screening use were similar for each geographic level: any modality was 61% for MSSA and 62% for county; FSIG was 20% for both levels.

Multilevel Assessment of CRC Screening

Table 2 shows multilevel logistic regression results for CRC screening use by any modality and by FSIG. Model 1 provides estimates of the between-area variation without any explanatory variables. In model 2, individual-level variables were added to determine their effects and to assess how much additional variation they explain. In model 3, MSSA and county variables were added to estimate their effects while controlling for individual-level characteristics and to assess whether they explain any additional variation.

By Any Modality. In model 1, the MORs are 1.43 and 1.18 for MSSAs and counties, respectively. These MORs indicate that the between-area variation was modest at the MSSA and small at the county level.50 Addition of individual characteristics in model 2 compared with model 1 further reduced the between-area variation for both MSSAs and counties as shown by the PCV: 21% for MSSA and 67% for county. Addition of area-level factors in model 3 did not change the between-area variation for either geographic level (PCV is 0% for both MSSA and county).

Results for individual-level characteristics in model 2 indicate that Latinos and Asian American and Pacific Islanders had statistically significantly lower odds of being up-to-date with CRC screening by any modality compared with whites. Those who were older, married/living with a partner, more educated, had higher income, and were insured had statistically significantly higher odds of being up-to-date.

In model 3, a statistically significant decrease in CRC screening odds was observed for MSSAs with a primary care shortage. The CRC screening uptake associated with individual characteristics observed in model 2 also held in model 3.

By Flexible Sigmoidoscopy. In model 1, the MORs are 1.55 and 1.37 for MSSAs and counties, respectively. These MORs indicate that the between-area variation is modest at the MSSA and county level.50 Addition of individual characteristics in model 2 compared with model 1 further reduced the between-area variation at the MSSA level (PCV = 29%) and resulted in more between-area variance at the county level (PVC = –45%). Addition of area-level factors in model 3 compared with model 2 did not explain additional between-area variance at the MSSA level (PCV = 0%), but did explain additional between-area variance at the county level (PVC = 31%).

Results for individual-level characteristics in model 2 indicate that African Americans were at significantly increased odds of being up-to-date with screening by FSIG compared with non-Hispanic whites, and American Indians/Alaska Natives and Asian American and Pacific Islanders were at significantly decreased odds of being up-to-date by FSIG. Those who were older, were married/living with a partner, had more education, had higher income, were insured, and did not face food insecurity were at significantly increased odds of being up-to-date with screening by FSIG.

In model 3, a statistically significant increase in CRC screening odds was observed for counties with a higher HMO penetration and a significant decrease in screening odds was observed for counties with a primary care shortage. Most of the CRC screening uptake associations with individual characteristics observed in model 2 also held in model 3 except the association for African Americans was no longer significant in model 3.

DISCUSSION

We used a multilevel framework with 2 geographic levels to examine CRC screening uptake in California. We found that area-level healthcare resources were important in shaping CRC screening use. Specifically, county-level healthcare resources, HMO penetration, and primary care shortages were important contextual factors associated with individual CRC screening use by FSIG. Further, we found that primary care shortage was significant for CRC screening by any modality at the MSSA level but not at the county level, emphasizing the importance of MSSAs for allocating healthcare resources. We also confirmed established associations between individual risk factors and CRC screening use, even when contextual variables were included in the models. Finally, we identified a new individual-level association: food insecurity.

County HMO penetration and primary care shortage were significant predictors of CRC screening by FSIG, possibly reflecting the lasting effect of HMO screening programs in California in the 1990s and early 2000s.21,34-37 Our study extends previous research on healthcare market characteristics to show that managed care activity is associated with cancer screening uptake.23,53-57 HMO penetration has been shown to be associated with CRC screening use by any modality in Asian American populations in California.23 We found this effect for California’s age-eligible population for screening by FSIG, but neighborhood poverty and other social factors were not associated. Smaller units of geography (eg, census tracts) may be more salient.58-60 This suggests that understanding the multilevel context that may influence CRC screening and other health behaviors requires great care in matching exposures to the appropriate level of geography.

At the individual level, we found that respondents reporting food insecurity were less likely to be up-to-date with CRC screening, which is consistent with our hypothesis that food insecurity is a competing priority or stressor that may be a barrier to accessing and utilizing CRC screening. This has not been extensively examined in previous studies of cancer screening. Our findings for race/ethnicity, income, education, sex, marital status, and insurance are consistent with previous literature.8,12,14,15,19,61-63 In our study, individual-level variables remained consistent, strong predictors even in the presence of contextual factors. Our results also parallel findings from 2 other studies that found individual-level factors explained more variation than contextual factors in CRC screening.23,30 The null associations for SES and segregation may have been due to the large set of individual-level factors that we were able to account for and include in our models, which may have mediated the associations between these 2 contextual factors and CRC screening use.

 
Copyright AJMC 2006-2019 Clinical Care Targeted Communications Group, LLC. All Rights Reserved.
x
Welcome the the new and improved AJMC.com, the premier managed market network. Tell us about yourself so that we can serve you better.
Sign Up