Fundamental Causes of Colorectal Cancer Mortality: The Implications of Informational Diffusion


Andrew Wang, Mailman School of Public Health, Columbia University, New York, NY 10032 (email:


Context: Colorectal cancer is a major cause of mortality in the United States, with 52,857 deaths estimated in 2012. To explore further the social inequalities in colorectal cancer mortality, we used fundamental cause theory to consider the role of societal diffusion of information and socioeconomic status.

Methods: We used the number of deaths from colorectal cancer in U.S. counties between 1968 and 2008. Through geographical mapping, we examined disparities in colorectal cancer mortality as a function of socioeconomic status and the rate of diffusion of information. In addition to providing year-specific trends in colorectal cancer mortality rates, we analyzed these data using negative binomial regression.

Findings: The impact of socioeconomic status (SES) on colorectal cancer mortality is substantial, and its protective impact increases over time. Equally important is the impact of informational diffusion on colorectal cancer mortality over time. However, while the impact of SES remains significant when concurrently considering the role of diffusion of information, the propensity for faster diffusion moderates its effect on colorectal cancer mortality.

Conclusions: The faster diffusion of information reduces both colorectal cancer mortality and inequalities in colorectal cancer mortality, although it was not sufficient to eliminate SES inequalities. These findings have important long-term implications for policymakers looking to reduce social inequalities in colorectal cancer mortality and other, related, preventable diseases.

Despite great medical advances in the prevention and treatment of the major chronic diseases causing the most deaths during the latter half of the twentieth century, socioeconomic disparities in mortality attributable to these conditions have substantially widened over time. One underlying cause of death that exemplifies this longitudinal patterning of death is colorectal cancer. In an effort to deepen our understanding of this perplexing epidemiologic trend, we elaborated on the so-called fundamental cause theory (FCT) (Link and Phelan 1995), including a novel focus on the diffusion of innovations. Specifically, we drew on FCT to suggest that our relatively new capacity to prevent colorectal cancer mortality has not been distributed equally throughout the U.S. population but instead has spread more rapidly to wealthier areas of the country. In addition, we used the observation that some areas tend to adopt innovations more rapidly than others, in order to investigate (1) whether this tendency helps explain how innovations diffuse to wealthier areas of the country; (2) how diffusion exerts its own independent influence on the distribution of mortality; and (3) the extent to which more rapid diffusion modifies the influence of socioeconomic status (SES) on aggregate death rates over time.


In the United States in 2008, 142,950 people were diagnosed with colorectal cancer, and 52,857 of them died from the disease (CDC 2012). Colorectal cancer mortality can be prevented through the removal of adenomatous polyps and the prompt treatment of cancer cells using effective radiation or chemo-based therapies (USCSWG 2009). Since approximately 1980, the spread and steady improvement of endoscopic and fecal-based screening techniques have made polyps and invasive cancer easier to detect at earlier stages of disease development. To date, colonoscopy is the best method for screening both distal and proximal colorectal cancer sites (Rex 2011), as well as for identifying advanced and nonadvanced adenomas in average-risk populations (Quintero et al. 2012). However, critiques of the existing screening techniques suggest that these approaches are not infallible but instead depend on histology, location in the colon or rectum, and quality of the endoscopic technique (Baxter et al. 2009). Thus, efforts to improve prevention methods, diagnostic capabilities, and subsequent treatment protocols are important to further reducing colorectal cancer mortality (Fletcher et al. 2010). The dissemination of effective screening is vital to prevention; moreover, modifiable risk factors have been identified that could lead to further reductions in the risk of developing colorectal cancer and serious sequelae (Joshu et al. 2011). Together, these factors have resulted in a consistent decline in colorectal cancer mortality beginning around 1980 and continuing until at least 2007 (Edwards et al. 2010).

Disparities in colorectal cancer mortality rates that favor relatively advantaged over relatively disadvantaged subpopulations in the United States have emerged or grown stronger since 1980 (Byers 2010; Marcella and Miller 2001). For instance, Wingo and colleagues (1998) found that compared with blacks, whites have better five- and fifteen-year survival rates at diagnosis. Similarly, Soneji and colleagues (2010) revealed that since effective screening and prevention became available, whites have experienced significantly greater reductions in mortality than blacks. There also are marked differences in colorectal cancer outcomes that are distributed according to socioeconomic status. For example, Boyd and colleagues (1999) noted that that the impact of SES on colorectal cancer survival has increased since 1980. In this article, we explore the recent emergence of socioeconomic disparities in colorectal cancer mortality and its relationship to informational diffusion in the United States between 1968 and 2008.

Fundamental Cause Theory

Fundamental cause theory (FCT) provides a useful context for understanding how socioeconomic inequalities in mortality emerge as overall age-adjusted mortality declines (Lutfey and Freese 2005). The theory posits that social inequalities in mortality are created, maintained, or exacerbated in part because of the multiple ways in which the benefits of biomedical knowledge and technology are accessed by and distributed among different segments of the population. Access to valuable resources allows individuals and groups to avoid newly identified risks as well as to adopt newly discovered protective factors, thereby preventing disease and death (Link and Phelan 1995; Link et al. 2008; Phelan and Link 2005; Phelan et al. 2004). This approach to examining and understanding social inequalities in health maintains that population-based disparities in morbidity and mortality arise when those with greater access to resources—including knowledge, money, power, prestige, and beneficial social connections—are better able to take advantage of effective innovations in disease prevention and treatment. It follows that as we develop new knowledge about prevention and treatment, the unequal distribution of that new knowledge leads to health inequalities because of SES. Because this can happen no matter what the risk and protective factors are in a given circumstance, the SES-mortality association tends to be reproduced in different places and at different times—a phenomenon that led Link and Phelan (1995) to call SES a “fundamental cause” of health inequalities.

Since its development, FCT has been tested, elaborated, qualified, and extended by sociomedical researchers (Link and Phelan 2010; Lutfey and Freese 2005; Miech et al. forthcoming; Phelan, Link, and Tehranifar 2010; Rubin, Colen, and Link 2010). In their discussion of health-related behavior, Link and colleagues demonstrated that individuals with more education became convinced of the causal relationship between cigarette smoking and lung cancer significantly earlier than did individuals with less education and that they were able to successfully engage in smoking cessation efforts based on this newly acquired knowledge (Link and Phelan 2009). Similarly, Chang and Lauderdale (2009) found that when statins emerged as an effective treatment for high blood cholesterol in the 1980s, there was a marked population-level inverse gradient in cholesterol levels according to household income, essentially reversing the observed association between SES and cholesterol in the 1970s. With respect to prevention, Link and colleagues (1998) noted that women with a higher SES were better able to access life-saving cancer screenings (e.g., Pap smears) than women with lower SES. Furthermore, the emergence of socioeconomic inequalities in morbidity and mortality once a disease becomes treatable is not unique to the United States. For instance, SES predicts similar disparities in breast cancer and stomach cancer screening rates in South Korea, in which higher SES was associated with increased screening and lower SES was associated with barriers to screening (Hahm et al. 2011).

Until recently, FCT provided only a partial explanation for the specific mechanisms by which SES inequalities in health and mortality change over time. Researchers have suggested that this theory and perspective can be temporally extended by examining longitudinal population-level mortality data (Miech et al. forthcoming). Clouston and colleagues (2011) posited that historical trends in mortality can be divided into four stages: (1) a natural-mortality stage in which prevention is not well understood; (2) an inequality-producing stage in which inequalities are created as individuals use resources to prevent mortality; (3) an inequality-reducing stage in which inequalities stabilize and may begin to decrease as beneficial knowledge or technology starts to saturate the population; and (4) a reduced-mortality stage in which inequalities and mortality rates stabilize or disappear altogether.

Saldana-Ruiz and colleagues (forthcoming) used this characterization to examine the impact of SES on colorectal cancer mortality, and they highlighted the influence of published recommendations of organizations like the American Cancer Society (ACS) and the U.S. Multi-Society Task Force on Colorectal Cancer. They showed that socioeconomic inequalities became apparent following the advances in colorectal cancer screening, which was initiated in 1980 along with the simultaneous publication of the ACS's first guidelines on preventing colorectal cancer through screening techniques. The guidelines from the ACS and the U.S. Multi-Society Task Force on Colorectal Cancer released between 1980 and 1989 recommended an annual digital rectal exam for those over forty years of age and an annual fecal occult blood test and a proctosigmoidoscopy for those over fifty years of age every three to five years. Saldana-Ruiz and colleagues (forthcoming) also suggested that even though socioeconomic inequalities in colorectal cancer mortality have only recently begun to emerge, they have increased substantially and that, if nothing is done to address these population-based health differentials, they are likely to diverge at an exponential pace.

Diffusion Processes

In considering the role of contextual factors in a fundamental cause framework, researchers have posited the concept of “meta-mechanisms” that may affect how social inequalities arise (Freese and Lutfey 2011). As described earlier, FCT emphasizes that risk and protective mechanisms change as we learn more about how to prevent disease and death. The concept of meta-mechanisms suggests that a set of more enduring linkages between SES advantage and health might be identified, thereby deepening our understanding of the association between SES and health. Freese and Lutfey (2011) identified four such meta-mechanisms: (1) the means by which individuals use SES-related resources to gain a health advantage; (2) spillovers, in which people benefit from being embedded in contexts in which others (e.g., spouses, neighbors, coworkers) are attentive to health concerns; (3) a health habitus, in which socially structured preferences for a “healthy lifestyle” shape health outcomes; and (4) institutions in which the focus is turned to the agency of institutions and their treatment of people from different socioeconomic backgrounds.

The concept of meta-mechanisms helps motivate our interest in geographic diffusion in the context of FCT. In 1964, Rogers developed diffusion of innovations theory to explain the spread of ideas, practices, and objects between individuals and organizations. Diffusion theory posits that following an innovation in prevention, new knowledge is passed through complex processes at all levels of social interaction, including individuals, institutions, and governments. To the extent that beneficial health knowledge, practices, and procedures are diffused by geographic area, the meta-mechanism of “spillovers” becomes salient because an individual will either benefit or be harmed, depending on whether that diffusion happens where he or she resides.

In this article, we examine state-level variation in the diffusion of innovations. Evidence that diffusion significantly impacts cancer outcomes is underscored by the research of Miller and colleagues (2008), who indicated that at least in the case of kidney cancer, the effective diffusion of information to physicians can be more important to the application of beneficial treatments than the patient's characteristics or disease etiology. Since the diffusion of information regarding the prevention and treatment of cancer operates on multiple levels, it can be hastened at the state level by increasing the capacity for individuals, including patients and health care providers, to learn and adopt beneficial health information, practices, and procedures. State-specific laws and policies have been shown to either increase or decrease the rate of diffusion (Karch 2006). Accordingly, speedier diffusion is likely to lead to a faster implementation of preventive information by all affected groups. Insofar as this implementation leads to reductions in mortality, it is likely that universally faster diffusion leads to more equal reductions.

Supporting a focus on the role of diffusion, one of the most remarkable findings to emerge from this field is a generally robust tendency for some geographic units to implement innovations at a much faster pace than others. For example, beta-blockers, an inexpensive and effective means of forestalling complications and death following a heart attack, were discovered in the 1980s. However, by 2000, the uptake of this life-saving innovation varied dramatically across the U.S. states and, remarkably, correlated 0.57 with the use of hybrid corn in those same states more than fifty years earlier (Skinner and Staiger 2005). This finding suggests not only that diffusion processes matter but also that relatively enduring features of places are important to shaping diffusion across seemingly unrelated innovations (e.g., hybrid corn and beta-blockers). Further theoretical support for the idea that diffusion occurs at both the aggregate and the individual level emerged from the work of Strang and Soule (1998), who argue that in order to understand how innovations are disseminated throughout a population, we should be more attentive to the role of social relations among individuals, media, experts, and government in influencing the adoption of new ideas and behaviors.

Therefore, one mechanism through which social inequalities might arise is the differential diffusion of information across macro-level contexts such as states or counties. For instance, with regard to breast and stomach cancer, inequalities in governmental intervention in South Korea have helped exacerbate differences in the diffusion and early adoption of preventive innovations (Hahm et al. 2011). Because the diffusion of information at the aggregate level can be influenced through directed policy, efforts to increase diffusion can substantially reduce social inequalities in mortality. Based on these insights, we employed a diffusion index developed by Walker (1969), which uses data from the 1960s to gauge state-level variations in the diffusion of innovations. We hypothesized that a state's tendency toward the faster diffusion of innovation (1) predicts state-specific declines in colorectal cancer mortality and (2) may modify the association between SES indicators and colorectal cancer mortality. If these hypotheses can be supported, it would suggest that the nature of the diffusion of knowledge and practices is a major, modifiable determinant of disparities in the population's health.

Data and Methods

Sample and Measures

To measure colorectal cancer mortality, we utilized data from detailed death records from 1968 to 2008 compiled by the National Center for Health Statistics (NCHS 2010). These data include county-specific counts of residents’ deaths, wherever they occur, that were attributable to colorectal cancer (ICD 8–9: 153–54; ICD 10: C18–C21). These counts are grouped into ten-year age groups and are further delineated by race and sex. We relied on midyear population counts to determine the population at risk. The data exclude deaths of noncitizens in the United States. Because very few people die from colorectal cancer before thirty-five years of age, we restricted our analyses to individuals who were thirty-five years or older at the time of death. We also exclude data from less than 1 percent of the counties owing to boundary changes between 1968 and 2008 that made those geographic areas inconsistent during our period of interest.

To capture the ways in which SES fluctuated over time, we constructed an aggregate measure of SES using county-level data from five decennial censuses. Values for the SES measure during intercensal years were obtained by means of linear interpolation. This indicator of SES is a modified version of the one used by Singh, Miller, and Hankey (2002) and is identical to the one employed by Rubin and colleagues (2010). The SES index is based on five county-specific measures, including the proportion of the population with fewer than nine years of education, with more than twelve years of education, without a phone, with a white-collar job, and living under the poverty line (αChronbach= 0.93). We used factor analysis to combine these variables and standardized the resultant SES measure. We also examined SES at the state level with the observed range at a minimum of −5.21 in Mississippi and a maximum of 4.81 in Connecticut.

To capture the rate at which new information and/or behaviors regarding disease prevention and treatment are likely to spread across different geographical areas, we employed an indicator of diffusion that was captured before our measures of SES, as well as colorectal cancer mortality and SES. Walker's influential analysis (1969) generated an “innovation score” to help explain “why some states adopt innovations more readily than others.” In his original study, Walker captured diffusion among forty-eight U.S. state legislatures by using data on the adoption patterns of eighty-eight social programs that were collected before 1965 and spanned the following twelve areas: welfare, health, education, conservation, planning, administrative organization, highways, civil rights, corrections and police, labor, taxes, and professional regulation. The resulting aggregate indicator varied from 0 (refusal of adoption) to 1 (complete and instantaneous adoption). In practice, the scores varied from a minimum of 0.298 (Mississippi) to a maximum of 0.656 (New York), illustrating the rate at which policy initiatives were adopted (hereafter, the propensity or tendency for faster diffusion). In the analyses presented here, we standardized the composite scores so that a unit increase refers to a one standard deviation increase in the rate of diffusion. This state-level measure assumes that innovation is uniform within each state. While we recognize that the variation within the state could be a desirable and important aspect to examine empirically, such information is not available to us at this time, leading us instead to harness the large variation in diffusion between states for our purposes.


We provided both descriptive and multivariate results to test our research ideas. For descriptive purposes, we used geographical maps using Microsoft MapPoint for both county-specific SES levels and state-specific propensity for diffusion (figure 1, panels A and B). We present the bivariate associations in the figures showing age-adjusted colorectal cancer mortality rates from 1968 to 2008, stratified first by county-level SES (figure 2) and then by state-level diffusion scores (figure 3). The yearly rates are highly variable, so in order to clarify the underlying historical process, we used Excel to overlay third-order polynomial trend lines. In figure 2, high SES refers to those counties that fall within the top tertile (33%) of this distribution, and low SES refers to counties that fall within the bottom tertile (33%) of this distribution. Similarly, in figure 3, a higher propensity for diffusion refers to states with overall diffusion scores in the highest tertile (33%), while slow diffusion refers to states with overall diffusion scores in the lowest tertile (33%). Graphical analysis, including maps and charts, uses county-level colorectal cancer mortality rates (per ten thousand) adjusted for age, race, and sex.

Figure 1.

Geographic Distribution of Main Independent Variables: (A) County-Level Gradations in Distribution of SES in 2000 and (B) State-Level Gradations in Propensity for Diffusion Before 1965.

Source: Data are from the authors' calculations for SES from data by the U.S. Census Bureau's Decennial Census; and diffusion from Walker 1969.

Figure 2.

Trends in Average Yearly Age-, Race-, and Sex-Adjusted Colorectal Cancer Mortality Rates, Separated into Tertiles of High, Middle, and Low Socioeconomic Status at the County Level, 1968–2008.

Source: Data are from the NCHS compressed mortality file, 1968–2008.

Figure 3.

Trends in Age-, Race-, and Sex-Adjusted Colorectal Cancer Mortality Rates, Separated into Tertiles of Fast, Medium, and Slow Diffusion at the State Level, 1968–2008.

Source: Data are from the NCHS compressed mortality file, 1968–2008.

We included the following statistical controls in multivariate regression models: age (ten-year age groups), race (black, white), sex (female, male), and urbanicity (the proportion of the population living in an urban area). To further examine the influence of time on our results, the data were aggregated into five-year intervals and entered as dummy variables into a subset of regression models in order to capture potential nonlinear changes in the association of SES, diffusion, and colorectal cancer mortality over time. Sensitivity analyses suggest that including a variable to assess year as an ordinal rather than a five-year categorical variable does not alter the conclusions we drew from the findings.

To extend these analyses into a multivariate context, we utilized negative binomial regression models on the original county-level mortality counts (Rubin, Colen, and Link 2010). Such strategies are appropriate when there is overdispersion in the dependent variable, a problem in which the variance is larger than the mean (Gardner, Mulvey, and Shaw 1995). Sensitivity analyses testing for zero inflation, a problem in which there are large numbers of observations with no deaths, showed no significant differences. Multivariate models incorporate SES and diffusion using the full range of continuous measures, rather than the more crude tertiles we used earlier for descriptive purposes. We calculated robust (Huber-White) standard errors, which were also clustered by county, and we used mortality rate ratios (MRRs) and 95 percent confidence intervals. MRRs provide risk ratios that compare the colorectal cancer mortality rate for one group using the reference category as denominator. For our two focal variables, SES and diffusion, we provided the MRR for a one standard deviation unit change from the sample mean (the reference), so that an MRR of 1.20 indicates a 20 percent increase in the risk of mortality associated with a one standard unit change in SES or diffusion, whereas an MRR of 0.80 indicates a 20 percent decrease in the risk of mortality.


Figure 1 presents two maps that depict the distribution of SES in 2000 and Walker's state-level diffusion scores. Panel A shows county-level SES distributions, and panel B shows state-level diffusion distributions. The highest SES counties are concentrated in the Northeast, the Midwest, across the northern and mountain states, parts of California, and the coasts of Florida, while Southern states are predominately, though not exclusively, in the lowest SES tertile. Of note, in panel B, which shows the state-level diffusion measure, the West Coast and the Northeast tend to be characterized as having a higher propensity for diffusion, while the South tends to be characterized by slower diffusion. Finally states with a higher propensity for diffusion (i.e., in the West and Northeast) also usually contain a number of high SES counties (in 2000, r = 0.37). However, a number of states, such as New Hampshire, could be characterized as having relatively high levels of SES but a tendency toward slower diffusion.

Figure 2 shows average colorectal cancer mortality rates and their trends over time by high, medium, and low levels of SES. Over the forty-year study period, colorectal cancer mortality rates steadily rose in the lowest SES counties. In contrast, since around 1980, colorectal cancer mortality rates have been declining in the highest SES counties. This trend is evident even though, compared with both low and middle SES counties, higher SES counties reported significantly higher rates of colorectal mortality at the beginning of the study period (1968–1970).

Figure 3 shows mortality rates stratified by tertiles of Walker's measure of state-level propensity for diffusion. States with a higher propensity for diffusion show substantial declines in colorectal cancer mortality over the forty-year study period, while those states with a propensity toward slow and moderate rates of diffusion experienced relatively small decreases in death rates due to colorectal cancer. For example, when compared with those states in the highest diffusion tertile, the states in the lowest diffusion tertile exhibited a decline in colorectal cancer mortality that began later and progressed at substantially slower rates, though the baseline mortality rates among those living in lower SES counties and slower-diffusing states were much lower on average than those in higher-diffusing states. By the early 2000s, mortality rates had declined in all states, regardless of the speed of diffusion.

SES can affect the likelihood of adopting innovations because communities with more social and economic resources can use them to adopt newly diffused innovations earlier. Table 1 provides negative binomial estimates for age, gender, and race when both SES and the rapidity of diffusion are included in the model. Model 1 provides baseline estimates; model 2 adds an SES interacted with time term and diffusion interacted with time term to model 1; and model 3 adds an SES interacted with diffusion term and an SES by diffusion by time interaction term to model 2.

Table 1. 
Negative Binomial Regression Estimates for Colorectal Cancer Mortality for Level (1) and Change (2) in SES and Diffusion and Their Interaction (3)
   Model 1  Model 2  Model 3  
  1. Note:*p < 0.05, **p < 0.01, ***p < 0.001. MRR = Mortality Rate Ratio, SE = Standard Error.

  2. Source: Mortality data from the Compressed Mortality File by the National Center for Health Statistics; authors’ calculations for SES from data by the U.S. Census Bureau's Decennial Census; and diffusion from Walker 1969.

Age Group45–5435–443.930.03***3.930.03***3.930.03***
 55–64 11.720.10***11.740.10***11.730.10***
 65–74 26.760.25***26.820.25***26.810.25***
 75–84 51.640.50***51.730.50***51.710.50***
 85+ 89.260.85***89.430.86***89.400.86***
Urbanicity  1.030.02 1.010.02 1.010.02 
Black  1.240.01***1.250.01***1.250.01***
Male  1.380.00***1.380.00***1.380.00***
 1976–80 0.960.01***0.990.01 0.990.01 
 1981–85 0.920.01***0.960.01***0.960.01***
 1986–90 0.870.01***0.930.01***0.930.01***
 1991–95 0.810.01***0.900.01***0.890.01***
 1996–00 0.750.01***0.840.01***0.830.01***
 2001–05 0.670.01***0.770.01***0.760.01***
 2006–08 0.610.01***0.700.01***0.690.01***
SES  1.020.01**1.080.01***1.090.01***
SES*Year1971–751968–70   0.990.01 0.990.01 
 1976–80    0.990.01 0.990.01 
 1981–85    0.990.01*0.990.01*
 1986–90    0.960.01***0.960.01***
 1991–95    0.930.01***0.920.01***
 1996–00    0.910.01***0.900.01***
 2001–05    0.890.01***0.880.01***
 2006–08    0.870.01***0.870.01***
Diffusion  1.050.00***1.100.01***1.160.01***
Diffusion*1971–751968–70   1.000.00 0.990.01*
 Year1976–80    0.980.00***0.970.01***
 1981–85    0.970.00***0.940.01***
 1986–90    0.960.01***0.920.01***
 1991–95    0.940.01***0.910.01***
 1996–00    0.930.01***0.890.01***
 2001–05    0.920.01***0.870.01***
 2006–08    0.910.01***0.860.01***
SES*Diffusion        0.950.01***
SES*Diffusion*1971–751968–70      1.010.00*
 Year1976–80       1.010.01*
 1981–85       1.020.01***
 1986–90       1.040.01***
 1991–95       1.040.01***
 1996–00       1.050.01***
 2001–05       1.060.01***
 2006–08       1.060.01***

Model 1 in table 1 shows the impact of age on colorectal cancer mortality, with each additional ten years of age conferring a substantial increase in risk of colorectal cancer mortality. Controlling for other factors in the model, urbanicity plays a marginally significant role in model 1, while both race and sex are strongly associated with an elevated risk of colorectal cancer mortality. Including SES and diffusion in model 1, we observe that a one standard deviation increase in SES was associated with a 2 percent average increase in the risk of colorectal cancer mortality and that a one standard deviation increase in propensity for diffusion was correlated with a 6 percent average increase in the risk of colorectal cancer mortality. These results suggest that over the entire period of observation, SES and diffusion were associated with higher mortality. The question to consider in model 2 is whether these associations changed in a predictable fashion over the period of observation.

Model 2 in table 1 shows that both SES (MRR = 1.08) and diffusion (MRR = 1.10) are associated with higher colorectal cancer mortality at the beginning of the observation period. However, model 2 demonstrates that the protective impact of both SES and diffusion emerged with time and with associated improvements in prevention and treatment. For SES, the protective impact grew from being nonsignificant in 1971–1980 (MRR = 0.99 for 1971–1975 and 1976–1980) to being highly significant in 2006–2008 (MRR = 0.87). This suggests that for each standard deviation increase in SES, people were 15 percent less likely to die of colorectal cancer in 2006–2008. Similarly, the protective impact of diffusion steadily grew from nonsignificance to 0.91 over the study period.

Model 3 of table 1 includes an interaction term between SES and diffusion over time. As evident in model 3 and the SES by time interaction, the impact of SES on colorectal cancer mortality is not lessened by the inclusion of the SES by diffusion interaction. But the interaction is significant and positive, indicating that when the propensity for faster diffusion is high, the impact of SES is attenuated. Figure 4 provides a visual explanation of these results. We examined the impact of SES on colorectal cancer mortality in three different types of states: (1) those with an average propensity for diffusion, (2) those with a propensity for diffusion one standard deviation above average, and (3) those with a propensity for diffusion one standard deviation below average. Diffusion has a moderating effect on the impact of SES over time: the effect of SES on colorectal cancer mortality is increased when diffusion is slower and is reduced when diffusion is faster. However, the impact of diffusion is not sufficient to eliminate the impact of SES.

Figure 4.

The Modified Impact of SES (MRRs) Estimated from Table 1, Assuming Three Different State-Level Diffusion Speeds, 1971–2008.
Note: This figure uses estimates from model 3 in table 1 to show the relative impact of higher SES scores in three scenarios: higher propensity for faster diffusion, compared with average diffusion and a propensity for slow diffusion. Significance is achieved at the 1981–1985 year group. The relative impact of a one standard deviation (1 SD) higher SES in counties with average diffusion, compared with states of 1 SD higher SES with diffusion 1 SD above average and diffusion 1 SD below average, with 95% confidence intervals (shown using bars).
Source: Data are from the NCHS compressed mortality file, 1968–2008.

Figure 4 uses estimates given earlier (table 1, model 3) to depict the impact of SES in three diffusion situations: one standard deviation above average diffusion, mean diffusion (table 1 shows the exact MRRs), and one standard deviation below average diffusion. Holding other factors constant, the impact of SES on colorectal cancer is substantially lower in faster-diffusing states than in slower-diffusing states. Specifically, the impact of SES is 0.92 in the fastest-diffusing state, whereas the influence of SES is more substantial in a state with an average propensity for diffusion (the MRR for SES is 0.87) and even more substantial in a state with a propensity for diffusion that is one standard deviation below average (0.82). This suggests that diffusion is important not only because it substantially reduces mortality (see table 1) but also because it helps reduce the influence of SES on colorectal cancer mortality.


We sought to understand the roles of SES and state-level diffusion in determining colorectal cancer mortality. We found that since the 1980s, the impact of SES on colorectal cancer mortality has risen substantially. We also found that those states with a propensity for faster diffusion also experienced faster reductions in mortality over this period. Finally, we found that the tendency for rapid diffusion had a moderating impact on the influence of SES: the influence of SES was attenuated in states with a higher propensity for faster diffusion. Our results support those of Saldana-Ruiz and colleagues (forthcoming), who noted that SES inequalities in colorectal cancer mortality have been increasing since the 1980s, and help fill a gap in the literature highlighted by Green and colleagues (2009), who suggested that researchers need to make a more concerted effort to understand the role of diffusion. The following discusses the importance of these findings to the literature and to future research.

With the introduction of more aggressive colorectal cancer screenings, better treatment protocols, and the publication of recommendations in the 1980s guiding that treatment, mortality attributed to colorectal cancer has significantly declined in the United States (Byers 2010). Despite these improvements, socioeconomic inequalities in colorectal cancer mortality rose during this same time period (Saldana-Ruiz et al. forthcoming). As noted earlier, FCT provides a means of understanding this perplexing pattern in epidemiological findings (Phelan et al. 2004). We found support for FCT, showing that SES inequalities in colorectal cancer mortality were produced following innovations in preventive information that were implemented in the 1980s. Moreover, following the innovation in colorectal cancer prevention and treatment, a faster diffusion of information was associated with a more rapid decrease in colorectal cancer mortality, along with SES. We also found, however, that contextual factors such as state-level informational diffusion modified the impact of SES on colorectal cancer mortality. Thus, our study helps fill a research need identified by Green and colleagues (2009), who argue that societal diffusion of information is a topic often overlooked in health research. That is, we have contributed to the growing literature examining the role of diffusion in health (Cunningham et al. 2012; Luke and Harris 2007) and have extended these analyses by considering the relevance of state-level diffusion to reducing SES inequalities in colorectal cancer mortality. Our research also provides a framework with which to understand the health implications of emerging technology. Innovations in screening and treatment, coupled with the greater effectiveness of existing treatments, increase the complexity of the information under consideration when treating patients. Since complex diagnoses can be difficult and complex guidelines can be misunderstood (Wegwarth et al. 2012), understanding the process of diffusion may further our understanding of the burden of colorectal cancer.

At the beginning of the study period, colorectal cancer mortality rates were higher in high SES counties and in fast-diffusing states, forming what is sometimes called a “reverse gradient” (Fernald and Adler 2008). Clouston and colleagues (2011) proposed that during a stage of natural mortality, when we have little knowledge of prevention, inequalities may be reversed for a number of reasons. Examples are better living conditions, which allow for survival long enough to die of colorectal cancer, or some unknown exposure related to both colorectal cancer outcomes and SES. We agree with Saldana-Ruiz and colleagues (forthcoming), who argue that the decrease in colorectal cancer is due in part to the role of recommendations and guidelines for effective screening and treatment. But Clouston and colleagues (2011) contend that SES inequalities might be exacerbated as people actively change to a healthier lifestyle in regard to our newfound knowledge. Edwards and colleagues (2010) support such a notion, clearly showing that behavioral changes are one way that people have been able to prevent the incidence of colorectal cancer. We thus propose that some aspects of the changing relationship between SES and colorectal cancer mortality may be due to behavioral changes.

Our results indicate that while SES inequalities in colorectal cancer mortality are increasing, more rapid diffusion can partly modify this relationship, thereby helping shape the distribution of death across a population. This analysis follows from a long history of studies on the complex ways that diffusion of information can work to affect mortality (Lomas 1993). Future research should clarify the sociostructural factors, including the influence of health care practitioners and social networks that depend on diffusion and modify its influence on colorectal cancer mortality. Some work in other forms of cancer may pave the way for such an analysis in relation to colorectal cancer. For example, Miller and colleagues (2008) find that compared with patients’ characteristics (such as age, sex, race, income, education, marital status, region, and tumor location), diffusion of surgical innovation among health care practitioners explains a larger proportion of the variance in outcomes related to two innovative kidney cancer treatments: radical partial nephrectomy and laparoscopic radical nephrectomy. Understanding the diffusion of innovation separate from the patient may not be sufficient in all cases, however. For instance, Finney Rutten, Nelson, and Meissner (2004) highlight the importance of diffusing information to patients in influencing the use of Pap smears, mammography, and colorectal cancer screenings. Future research should consider these analyses in light of information about both individual- and practitioner-level measures of informational diffusion.

Other Cases

Our results may have implications for other causes of death, for instance, lung cancer. Lung cancer mortality can be prevented by avoiding tobacco or not smoking, a determinant that has been publicized since the mid-1950s (Alberg and Samet 2003). The link between smoking and lung cancer may not have been universally diffused, however. Earlier research showed that mortality from lung cancer depends on SES over time, suggesting that SES inequalities grew substantially after the publication of the link between smoking and lung cancer, even though lung cancer mortality was still on the rise (Krieger et al. 2012; Miech et al. forthcoming; Rubin, Colen, and Link 2010). But little is known about whether more rapid diffusion might lead to reduced mortality and reduced SES inequalities in mortality over time. Future research should consider how the rapid diffusion of information modifies SES inequalities in lung cancer mortality.


This study has a number of limitations. First, our area-based data did not permit us to analyze the relationship between SES and colorectal cancer at the individual level, including the proximate mechanisms that are likely to explain this robust association. We believe, however, that this analysis would be both useful and necessary, for two overarching reasons. First, a major tenet of FCT is that SES is predictive of inequalities in mortality, regardless of the mediating pathways or the aggregate level. Second, colorectal cancer mortality is relatively rare, and thus any nationally representative sample is unlikely to have sufficient mortality to allow both historical and individual-level analysis. Nevertheless, future studies considering the historical processes determining colorectal cancer mortality should attempt to incorporate person-specific information as well as data on the social contexts that they inhabit.

A second limitation is our reliance on multiple years of death certificate data, during which time the ICD codes changed. This would bias our results if these changes were systematically related to county-level SES or rates of diffusion. Given that changes to ICD codes are initiated and instituted at the national level and diffuse relatively rapidly throughout the states and counties, the sources of reporting bias are likely to be random.

A third limitation concerns our decision to focus on understanding how colorectal cancer mortality is distributed by SES but not by race. Our decision for race to be a control variable was deliberate. One reason is that a number of analyses have convincingly shown substantial racial inequalities in colorectal cancer outcomes (Hao et al. 2011; Hill et al. 2010; Marcella and Miller 2001), so we decided not to undertake a detailed analysis of racial inequalities.

Finally, we used diffusion categorizations that were defined by Walker (1969)before we measured mortality, so as to reduce the potential for reverse causation. Carter and LaPlant (1997) more recently analyzed how adding more information and following up to the mid-1990s might change this categorization and found that even though more than half the states remained similarly ranked, some change was evident. This change may lead to conservative estimates if states change their amenability to diffusions of innovation over time or if states are more likely to adopt certain types of programs more than others (for an analysis of the predictors of civil liberty legislation diffusion, see Vasi and Strang 2009). Those acting as role providers in adopting and understanding screening guidelines often poorly understand recommendations (Wegwarth et al. 2012), which slows the diffusion and implementation of preventive information and adds another layer of complexity. Acknowledging these limitations, we next discuss the larger implications of this research.

Policy Perspectives

Health policies globally have acknowledged that risks and exposures are structured by social factors, including SES at the city, state, and national levels (Wilkinson and Marmot 2003). But health policy is not relegated to the health care field but includes a variety of social, educational, informational, and economic policies that affect both the distribution and the efficiency of health care interventions (Kaplan and Lynch 2001). Link (2008) provides a mechanism for the influence of this societal inequality and policy by suggesting that societies “create and shape” their experiences of health and illness by promoting such things as the diffusion of information and increasing equitable access to care.

Our results highlight the influence of informational diffusion with innovation and how it may shape the reduction of inequalities in colorectal cancer at both the individual and the population level. We thus present evidence in concurrence with Graham (2004), suggesting that policy agendas must be in place to ensure reductions in social influences on health and social inequalities by accounting for the diffusion of information as an integral part of that process. Our findings therefore support Greenhalgh and colleagues (2004) and imply an important point of intervention: that increasing the propensity for faster diffusion of information at all levels of society, including patients and providers, may have a large impact on reducing both colorectal cancer mortality and the social inequalities observed in colorectal cancer mortality.


By examining geographical and contextual factors, we confirm the importance of fundamental cause theory in determining how social disparities in colorectal cancer arise. Socioeconomic inequalities in colorectal cancer mortality are increasing substantially and are likely to continue to grow over time. Our analyses are therefore both novel and timely because they demonstrate that diffusing relevant information and innovation may both help protect individuals from mortality and reduce social inequalities in mortality. Further reductions in colorectal cancer mortality are possible although efforts to reduce mortality may require a greater knowledge of the factors that lead to large health disparities. In order to maximize these reductions, future efforts in health policy need to focus on increasing both the diffusion of and the access to new medical information while also seeking to reduce social inequalities resulting from that information.


Acknowledgments: This work was funded by the Centers for Disease Control and the Canada / U.S. Fulbright Program. The authors also thank the Robert Wood Johnson Foundation Health and Society Scholars program for its financial support. This project was carried out at Columbia University's Mailman School of Public Health.