Effects of Text Cohesion on Comprehension and Retention of Colorectal Cancer Screening Information: A Preliminary Study

Abstract

Increasing readability of written cancer prevention information is a fundamental step to increasing awareness and knowledge of cancer screening. Instead of readability formulas, the present study focused on text cohesion, which is the degree to which the text content ties together. The purpose of this study was to examine the effect of text cohesion on reading times, comprehension, and retention of colorectal cancer prevention information. English-speaking adults (50 years of age or older) were recruited from local communities. Participants were randomly assigned to read colorectal cancer prevention subtopics presented at 2 levels of text cohesion: from higher cohesion to lower cohesion, or vice versa. Reading times, word recognition, text comprehension, and recall were assessed after reading. Two weeks later, text comprehension and recall were reassessed. Forty-two adults completed the study, but five were lost to follow up. Higher text cohesion showed a significant effect on reading times and text comprehension but not on word recognition and recall. The effect of text cohesion was not found on text comprehension and recall after 2 weeks. Increasing text cohesion facilitates reading speed and comprehension of colorectal cancer prevention information. Further research on the effect of text cohesion is warranted.

Colorectal cancer (CRC) is among the most common cancers affecting both men and women worldwide, especially in economically developed countries (Parkin, Bray, Ferlay, & Pisani, 2005). Although CRC incidence and mortality rates are slowly declining in the United States (Edwards et al., 2010), it remains the third leading cause of cancer death (Jemal, Siegel, Xu, & Ward, 2010). The majority of CRC cases and deaths can be prevented through lifestyle change and screening tests (Huxley et al., 2009; U. S. Preventive Services Task Force, 2002). Current screening guidelines recommend adults at average risk start screening at age 50 years: having a fecal occult blood test/fecal immunochemical test each year, a flexible sigmoidoscopy or double-contrast barium enema every 5 years, or a colonoscopy every 10 years (Levin et al., 2008). However, studies have reported consistent underutilization of CRC screening tests (Mitka, 2008; Shapiro et al., 2008).

Lack of knowledge about cancer and screening tests is a common barrier to CRC screening (Ford, Coups, & Hay, 2006; Shapiro et al., 2008; Wee, McCarthy, & Phillips, 2005). Increasing the public's knowledge about CRC and benefits of screening is an important adjunct to physicians' recommendation of cancer screening. Cancer prevention campaigns aim to increase screening rates through education about CRC risk factors, symptoms, and screening tests. Many campaigns use written information, in electronic or print form, to disseminate cancer prevention information. However, low readability of written information creates a barrier to understanding (Friedman, Hoffman-Goetz, & Arocha, 2004; Kaphingst, Zanfini, & Emmons, 2006; Singh, 2003). The readability of written health information is often measured by formulas, such as the Flesch-Kincaid Grade Level, the Gunning Fog Index, the Fry Readability Graph, and the Simple Measure of Goobledygook index (Ley & Florio, 1996). These formulas measure word and sentence lengths and indicate readability in terms of reading grade levels. Cancer prevention information is mostly written above the 10th-grade reading level (Friedman & Hoffman-Goetz, 2006; Friedman et al., 2004; Kaphingst et al., 2006; Singh, 2003), exceeding the 8th-grade reading level of most American adults (Kirsch, Jungeblut, Jenkins, & Kolstad, 1993).

Although these formulas have been widely used to measure readability, their utility is limited (Bailin & Grafstein, 2001; Davision & Kantor, 1982; Hoar & Hoar, 1981; Kintsch & Vipond, 1979; Klare, 1963; Meadea & Smith, 1991; Pichert & Elam, 1985; Rubin, 1985). One common critique of readability formulas is that they overlook crucial text variables and are not grounded in comprehension theory (Bailin & Grafstein, 2001; Davision & Kantor, 1982; Klare, 1963; Rubin, 1985). Readability formulas do not reflect organization of content; that is, they do not distinguish between organized sentences and scrambled ones. Texts with low reading grade levels may not be necessarily highly readable or understandable, especially if text cohesion is low. The degree of text cohesion is not measured by commonly used readability formulas.

Text cohesion is defined as “relations of meaning that exist within the text, and that define it as a text” (Halliday & Hasan, 1976, p. 4). Contemporary comprehension theory suggests that text cohesion guides people as they process text information while reading (Kintsch & van Dijk, 1978). Text cohesion reflects a deeper level of comprehension process (Kintsch & Vipond, 1979). During reading, meaning is abstracted from lower levels of text representation to form higher levels of text representation (Radvansky, 1999). Deriving meaning from text follows a vertical direction from lower levels to higher levels, and is also cyclic, alternating between construction and integration phases (Kintsch, 1998). Ideas that readers extract from the lower level are linked to each other as well as to ideas from earlier cycles to form a network. The links' relevance to each other determines their strength. As a result of the cyclical process, only strong links are kept in working memory for comprehension. Text cohesion plays an important role at the higher levels of text processing in the path to comprehending written information. For adult readers, text cohesion may be more essential to comprehension than word difficulty and sentence length.

Text that is highly cohesive maintains continuity of ideas. If there are few or no connections between ideas/sentences in a given text, readers need to bridge the cohesion gap through inferences (Singer & Ritchot, 1996). Comprehension suffers when readers are not able to make such inferences, especially when they have little or no relevant background knowledge on which to draw. For example, someone with limited knowledge of CRC risk factors may not fully understand the following sentences because there is low cohesion between the first and second sentences:

Cancer of the colon and rectum happens to men and women. Talk to your family. Ask your grandparents, parents, sisters or brothers. Ask if they know if they have ever had a polyp, colorectal cancer, a bowel disease or some other type of cancer. Tell your doctors if you or anyone in your family has had these problems.

Health information written with low text cohesion is generally more difficult to understand when readers do not have domain-specific knowledge to make necessary inferences. This can be the case with CRC prevention information as awareness of this cancer in the general public is low (Ford et al., 2006).

Text cohesion arises from the use of explicit features (e.g., words, phrases, sentences) that make connections among ideas and sentences to guide the reader through the text. The present study focused on two types of text cohesion—referential cohesion and semantic cohesion—because research has found that both account for the greatest variance of text cohesion in health texts (Liu, Kemper, & Bovaird, 2009). Referential cohesion means that a noun, pronoun, or noun phrase refers to another constituent in a text. Referential cohesion can be increased by repeating prior arguments, that is, nouns, pronouns, noun-phrases (Vidal-Abarca, Martinez, & Gilabert, 2000). The noun, “colon,” is an example of referential cohesion in the following two sentences: “The colon and rectum are parts of the digestive system. The colon is where waste material is stored.” Semantic cohesion is the semantic, or conceptual, similarity of two sentences or paragraphs. Semantic cohesion can be increased by enhancing semantic connections of two text constituents, such as words, phrases, or sentences, that share the same world knowledge (Kintsch, 1998). Consider the following explanation of sigmoidoscopy as an example. ” … A doctor puts a flexible viewing tube into your rectum and into the first part of your colon. This lets the doctor see the lower portion of the intestine, which is where most colon cancers grow ….” There is a semantic connection between the term “rectum” and the phrase “the first part of your colon” in the first sentence and the phrase “the lower portion of the intestine” in the second sentence. There is also a semantic connection between the word “viewing” in the first sentence and the word “see” in the second sentence. These textual features help readers construct a knowledge-based representation of sigmoidoscopy and make the information easier to comprehend. In contrast, the same example with lower semantic cohesion is ” … A doctor puts a flexible viewing tube into the lower part of the intestine. This allows the doctor to examine your colon for ….”

Health literacy is defined as “the degree to which individuals have the capacity to obtain, process, and understand basic health information and services needed to make appropriate health decisions” (Committee on Health Literacy & Board on Neuroscience and Behavioral Health, 2004). CRC prevention information written with high text cohesion should be easier to process and understand, and therefore, increase the likelihood a person will act on the information by getting a screen test. The purpose of the present study was to examine whether CRC prevention information written with high text cohesion is easier to read, understand, and remember than information written with lower text cohesion. We have developed a conceptual framework (see Figure 1) to guide this study on the basis of contemporary comprehension theory and a review of the literature. As the figure illustrates, outcomes of CRC prevention information depend on text readability and also readers' characteristics. Understanding and retention of CRC prevention information will increase readers' domain knowledge of CRC. Highly cohesive text requires less inferential reasoning to understand, so reading time would be reduced and comprehension would be improved. Highly cohesive text should also help readers encode information more deeply in memory resulting in better retention. Compared with CRC prevention information with low text cohesion, we hypothesized that information with high text cohesion would be (a) easier to read by showing shorter reading times; (b) easier to understand by showing better text comprehension; (c) and easier to remember by showing better word recognition, and better recall.

Figure 1 Conceptual framework.

Method

Development of CRC Prevention Texts

To select experimental texts for this study, we searched websites and local hospitals for CRC prevention information. We used the search terms colorectal cancer, colon cancer, and cancer research center in four online search engines: Google, Yahoo!, MSN Life Search (Bing), and Ask.com. The first 20 websites resulting from each search were reviewed, along with print materials obtained from local hospitals. Materials that did not cover the following four topics were excluded: (a) locations of the colon and rectum, (b) symptoms of CRC, (c) risk factors for CRC, and (d) CRC screening tests. We analyzed 30 texts using Coh-Metrix, an online linguistic and discourse computational tool for analyzing text cohesion (Graesser, McNamara, Louwerse, & Cai, 2004). The Coh-Metrix provides various indices of text cohesion. A composite score indicating the degree of referential cohesion and semantic cohesion was computed for each text. Two texts, one written with high cohesion—10th percentile of the 30 composite scores, and the other written with low cohesion—90th percentile of the 30 composite scores, were mixed and modified to create the two experimental texts for this study. The reading grade levels of both texts were comparable (seventh grade) after minor revision. In one text, text cohesion was high in the first two subtopics (locations of the colon and rectum, symptoms of CRC) but low in the last two subtopics (risk factors for CRC, CRC screening tests). In the other text, text cohesion was low in the first two subtopics (locations of the colon and rectum, symptoms of CRC) but high in the last two subtopics (risk factors for CRC, CRC screening tests). Experimental text with high cohesion has more argument overlaps and semantic connections.

Research Design

We used a 2 × 2 cross-over design. Text cohesion (high versus low) was a repeated-measures within-subject effect. To prevent carryover effects, high-cohesion text content and low-cohesion text content described different information about CRC. Participants were randomly assigned to Group A or Group B. In Group A, participants read CRC prevention text with the first two subtopics written with high cohesion and the last two subtopics written with low cohesion. Participants in Group B read CRC prevention text with the first two subtopics written with low cohesion and the last two subtopics written with high cohesion. Text excerpts by group assignment are presented in Table 1.

Table 1. Group assignment and colorectal cancer prevention information

Group A	Group B
Subtopic 1: Locations of the colon and rectum.	Subtopic 1: Locations of the colon and rectum.
Text cohesion level: High	Text cohesion level: Low
“The colon is the part of the digestive system. The colon is where the waste material is stored. The rectum is the end of the colon adjacent to the anus. Together, the colon and rectum form a long muscular tube … .”	“The colon and the rectum are parts of the digestive system. Together, they form a long muscular tube called the large intestine. It also called the large bowel. The colon is the upper five to six feet of the large intestine … .”
Subtopic 2: Symptoms of colorectal cancer.	Subtopic 2: Symptoms of colorectal cancer.
Text cohesion level: High	Text cohesion level: Low
” … Colon cancer can be present for several years. The cancer can be present before symptoms develop. Symptoms vary according to where in the large bowel the tumor is located. The right colon is spacious. Cancers of the right colon can grow to large sizes before they cause any abdominal symptoms … .”	“There may be no symptoms with colorectal cancer. Warning signs to watch for may include: A change in bowel habits. Blood in, or on the stool (either bright red or very dark in color). Stools that are narrower than usual. Rectal bleeding … .”
Subtopic 3: Risk factors for colorectal cancer.	Subtopic 3: Risk factors for colorectal cancer.
Text cohesion level: Low	Text cohesion level: High
“Polyps. The vast majority of colorectal cancers develop in polyps. They are benign but they may become cancerous over time. Removing polyps is an important way to prevent colorectal cancer. Age. Colorectal cancers most often occur in people over the age of 50 … .”	“Some people are more likely to develop colorectal cancer than others. Factors that increase a person's risk of colorectal cancer include …. The presence of polyps in the large intestine. Most colon cancers develop in colon polyps. Therefore, removing benign colon polyps can prevent colorectal cancer … The extra tissue is called polyps. Colon polyps are initially benign. Over years, benign colon polyps can become cancerous.”
Subtopic 4. Colorectal cancer screening tests.	Subtopic 4. Colorectal cancer screening tests.
Text cohesion level: Low	Text cohesion level: High
” … Fecal Occult. Beginning at age 50, have a fecal occult blood test every year. This test checks for hidden blood in the stool. It is done because colorectal cancer may cause bleeding that cannot be seen. Sigmoidoscopy … .”	” … Fecal occult blood test. An important screening test for colorectal cancers and polyps is the stool occult blood test. Individuals over the age of 50 should have their stool tested for hidden or “occult” blood. Stool samples can be obtained to test for occult blood. Tumors of the colon and rectum tend to bleed slowly into the stool. The small amount of blood mixed into the stool is usually not visible to the naked eye … .”

Participants were recruited through word of mouth or research flyers posted in local churches, grocery stores, community centers, and senior centers. Inclusion criteria were as follows: (a) 50 years of age or older; (b) English speaking; and (c) access to a telephone. Data were collected during one in-person visit and one telephone follow-up interview at 2 weeks. Eligible participants completed the in-person visit at a university research lab or a private room in a senior center. Participants received a $20 gift card after the in-person visit and a $10 gift card after the telephone follow-up interview. The Indiana University Institutional Review Board approved this study before implementation.

Procedure

Written informed consent was obtained during the in-person visit. Demographic information, experience with CRC screening tests, and prior knowledge of CRC were collected through a structured interview. After the interview, participant completed the short version of the Test of Functional Health Literacy in Adults (Nurss, Parker, Williams, & Baker, 2001) and the Shipley vocabulary test (Shipley, 1940) sequentially. Next, the participant read the CRC prevention information.

The information was presented one sentence at a time on a 17-inch computer monitor. After reading each sentence, the participant pressed the space bar to continue reading. This line by-line sentence presentation was used to ensure that participants read each sentence and could not reread prior text so cognitive process was involved in building a cohesive representation of the text. Before reading the CRC prevention information, participants read a practice text to help them become familiar with the computer. After reading the CRC prevention information, participants completed a word recognition test and then a text comprehension test. Last, they completed two working memory tests and a recall test. Responses to recall questions were audiotaped for later scoring. A 2-week telephone follow-up interview was scheduled at the end of the visit. During the follow-up interview, participants first completed the recall test followed by the text comprehension test. Follow-up interviews were audiotaped for later scoring.

Measures of Readers' Characteristics

Prior Knowledge of CRC

Prior knowledge of CRC risk, symptoms, and screening tests was assessed with 10 multiple-choice questions. The questions were created by a behavioral oncology specialist (See the Appendix). All questions and response choices were read to the participants. Total knowledge scores were computed by a count of correct answers.

Health Literacy

Health literacy was measured using the short version of the Test of Functional Health Literacy in Adults (Nurss et al., 2001). The test is a 36-item Cloze test. Participants chose words from four possible answers to complete sentences in two health texts. The test has a reliability of .97 (Cronbach's alpha) and a correlation of .81 with another well-established health literacy test (Baker, Williams, Parker, Gazmararian, & Nurss, 1999).

Vocabulary Knowledge

Vocabulary knowledge was measured using the 40-item multiple choice Shipley vocabulary test (Shipley, 1940). Participants were instructed to choose the word with the most similar meaning to a given word from a list of four possible answers. The range of scores is 0–40, with higher scores indicating better vocabulary knowledge. The reliability of the test was reported at .87 (Shipley, 1940).

Working Memory

Digits Forward and Digits Backward tests from the Wechsler Adult Intelligence Scale were used to assess working memory (Wechsler, 1958). Participants were required to recall a series of digits in forward and backward order. The length of digits was increased gradually until participants failed two trials with the same digit length. Higher scores indicate larger working memory capacity.

Outcome Measures

Reading Time

Reading time for each sentence was recorded during reading. To adjust for variability in word length and text length among subtopics, reading time was converted to millisecond per letter.

Text Comprehension

We used 16 yes–no questions developed by the investigators to assess comprehension of common content in both experimental texts. For example, can a change in bowel habits be a symptom of colon cancer? The purpose of using yes–no questions was to measure participants' comprehension of factual information in the CRC prevention text. Four questions covered each subtopic; therefore, eight questions covered subtopics 1 and 2 and another eight covered subtopics 3 and 4. The text comprehension score is equal to the count of correct answers. High-cohesion content (subtopics 1 and 2 in Group A; subtopics 3 and 4 in Group B) and low-cohesion content (subtopics 3 and 4 in Group A; subtopics 1 and 2 in Group B) were scored separately for each participant.

Word Recognition

A word recognition test developed by the authors was used to evaluate participants' retention of words read in the CRC text. Word recognition measures lower-level text representation. Twenty-four words that appeared in both CRC prevention texts were selected as targets. Half of the words were from subtopics 1 and 2, and the other half were from subtopics 3 and 4. An additional 16 words that never appeared in the experimental texts were used as foils. Words were presented one at time in random order on a computer screen through EPRIME software (Schneider, Eschman, & Zuccolotto, 2002). Shortly after reading, participants were asked to judge whether or not they had read each word earlier. Word recognition scores were computed by summing the number of target words accurately recognized.

Recall

Participants' ability to retain CRC prevention information was also evaluated through a recall test. The recall test was used to measure higher level text representation. One probing question was provided for each subtopic. One point was given for each piece of information correctly recalled and only half of a point was given when the information was recalled partially. Only common information appeared in both texts was scored. High-cohesion content and low-cohesion content were scored separately in each group.

Statistical Analysis

Linear mixed model analyses, using SAS software, were conducted to compare the effect of text cohesion on reading times, word recognition scores, and text comprehension scores during the in-person visit as well as during the 2-week follow-up. The mixed model included the level of text cohesion (high vs. low; i.e., the treatment effect), subtopic (subtopics 1 and 2 vs. subtopics 3 and 4), and the interaction of both. Although to test subtopic effect was not the purpose of this study, the subtopic effect is generically referred to as the period effect in cross-over analyses so it needs to be accounted for in the analyses. Participants were treated as random effects in the model. The within-subject text cohesion comparison from the mixed model was not suitable to analyze recall results because subtopics 3 and 4 contained more information than subtopics 1 and 2. To examine the effect of text cohesion on recall, recall results were analyzed using the t test. All tests were performed at the .05 significance level.

Results

Forty-two people completed the in-person visit. Five were lost to follow-up resulting in 37 who completed the 2-week follow-up. Twenty participants (47.6%) reported having had a CRC screening test within the recommended timeframe, making them adherent to CRC screening guidelines. Participants, in general, had adequate health literacy but modest scores on prior knowledge of CRC and screening. Table 2 shows the descriptive statistics of demographic data, health literacy, vocabulary knowledge, working memory, and prior knowledge of CRC and screening by group. There were no statistically significant differences between the two groups on these variables.

Table 2. Participant characteristics, by group

	Participants who completed the in-person visit
Variable	Group A (n = 19)	Group B (n = 23)	Total (n = 42)
Sex, n
Male	5	12	17
Female	14	11	25
Race/ethnicity, n
Caucasian	8	11	19
African American	11	12	23
Health literacy, n
Less than adequate	0	1	1
Adequate	19	22	41
	M (SD)	M (SD)	M (SD)
Age in years	61.95 (8.53)	63.17 (8.58)	62.62 (8.47)
Years of education	13.05 (2.17)	12.65 (2.21)	12.83 (2.17)
Vocabulary score	27.53 (4.60)	26.65 (6.02)	27.05 (5.38)
Working memory ^a	6.05 (1.79)	6.32 (1.38)	6.20 (1.60)
Prior knowledge	5.42 (1.61)	5.70 (1.71)	5.57 (1.66)

Results of Reading Times, Text Comprehension, and Word Recognition

Table 3 shows descriptive statistics for reading times, text comprehension, and word recognition scores by group and by text cohesion level. Table 4 shows results from the linear mixed models analyses. Reading times, word recognition scores, and text comprehension scores were outcomes in separate models. Significant main effects of text cohesion and subtopic were found on reading times. Participants read subtopics written with high text cohesion faster (M = 98.51, SD = 33.76) than subtopics written with low text cohesion (M = 110.47.51, SD = 39.34), p = .010. Participants also read subtopics 3 and 4 faster (M = 94.33, SD = 29.78) than subtopics 1 and 2 (M = 114.64.51, SD = 40.77), p < .001. Participants did not have higher comprehension scores on subtopics written with high cohesion (M = 6.24, SD = 1.41) than those written with low cohesion (M = 6.17, SD = 1.42) at the in-person visit. Similarly, they did not have higher comprehension scores on subtopics written with high cohesion (M = 5.53, SD = 1.45) than those written with low cohesion (M = 5.76, SD = 1.34) at the 2-week follow-up. There was also no difference in recognizing target words between subtopics written with high text cohesion (M = 10.07, SD = 1.37) and subtopics written with low text cohesion (M = 10.36, SD = 1.48). However, participants recognized more target words from subtopics 3 and 4 (M = 10.98, SD = 1.00) than subtopics 1 and 2 (M = 9.45, SD = 1.38), p < .001.

Table 3. Means and standards deviations of reading time, text comprehension, word recognition, and recall scores

Group	Subtopic	Text cohesion level	Variable	n	M	SD
Group A	Locations of the colon and rectum (1), and symptoms of colorectal cancer (2)	High	Reading time (millisecond per letter)	19	111.18	38.95
			Text comprehension score: in-person visit	19	6.37	1.26
			Text comprehension score: follow-up	18	5.33	1.14
			Word recognition score	19	9.53	1.50
			Recall score: in-person visit	19	4.00	1.67
			Recall score: follow-up	18	2.83	1.29
	Risk factors for colorectal cancer (3), and colorectal cancer screening tests (4)	Low	Reading time (millisecond per letter)	19	101.72	33.58
			Text comprehension score: in-person visit	19	6.11	1.05
			Text comprehension score: follow-up	18	5.72	1.41
			Word recognition score	19	10.89	0.94
			Recall score: in-person visit	19	7.16	2.57
			Recall score: follow-up	18	4.94	1.70
Group B	Locations of the colon and rectum (1), and symptoms of colorectal cancer (2)	Low	Reading time (millisecond per letter)	23	117.39	42.79
			Text comprehension score: in-person visit	23	6.22	1.73
			Text comprehension score: follow-up	19	5.80	1.32
			Word recognition score	23	9.39	1.31
			Recall score: in-person visit	23	3.71	1.42
			Recall score: follow-up	19	2.89	1.76
	Risk factors for colorectal cancer (3), and colorectal cancer screening tests (4)	High	Reading time (millisecond per letter)	23	88.48	25.62
			Text comprehension score: in-person visit	23	6.13	1.55
			Text comprehension score: follow-up	19	5.70	1.69
			Word recognition score	23	11.04	1.07
			Recall score: in-person visit	23	7.57	3.19
			Recall score: follow-up	19	6.05	2.82

Table 4. Linear mixed model results in reading time, word recognition, and text comprehension scores

Outcome	Predictor	Parameter estimate	Parameter standard error	F	df	p
Reading time (millisecond per letter)	Text cohesionSubtopicText cohesion by subtopic	−13.2415.677.03	10.9810.9820.73	7.2228.090.11	414141	.010*< .001**.736
Text comprehension score: in-person visit	Text cohesionSubtopicText cohesion by subtopic	0.030.110.13	0.450.450.78	0.170.660.03	414141	.685.422.873
Text comprehension score: follow-up	Text cohesionSubtopicText cohesion by subtopic	–0.020.08–0.44	0.460.460.82	1.400.490.30	363636	.244.488.590
Word recognition score	Text cohesionSubtopicText cohesion by subtopic	–0.151.50–0.01	0.380.380.65	0.5461.610.00	414141	.465< .001**.983

Post Hoc Analyses of Text Comprehension and Word Recognition Scores

The aforementioned result of reading times suggests that participants read subtopics written with high text cohesion faster than subtopics written with low text cohesion. Participants allocated more time reading low-cohesion text because they might spend time bridging cohesion gaps in order to comprehend the text. To account for the tradeoff between reading times and text comprehension as well as reading times and word recognition, the reading times were divided by the count of correct answers in the text comprehension test and also the word recognition test. Linear mixed model analysis showed a significant effect of the text cohesion on text comprehension during the in-person visit (F = 7.07, p = .011). In other words, the average reading time per letter required for each correct answer was shorter for high-cohesion text (M = 17.01, SD = 9.05) than for low-cohesion text (M = 20.09, SD = 12.97). There was no statistical difference between text cohesion on word recognition (F = 3.15, p = .083). Namely, the average reading time per letter required for each correctly recognized word was similar for high-cohesion text and for low-cohesion text. A significant main effect of subtopic was found on word recognition (F = 48.43, p < .001) as well as text comprehension, both at the in-person visit (F = 7.27, p = .010) and the follow-up (F = 6.81, p = .013). The average reading time per letter required for each correct answer for subtopics 3 and 4 was shorter than for subtopics 1 and 2 on both tests.

Results of Recall

Descriptive results for recall scores were reported in Table 3. There was no difference in subtopics 1 and 2 between Group A (high text cohesion) and Group B (low text cohesion) during the in-person visit, t(38) = .59, p = .56, and follow-up, t(35) = .12, p = .91. Similarly, there was no difference in subtopics 3 and 4 between Group A (low text cohesion) and Group B (high text cohesion) during the in-person visit, t(38) = .45, p = .66, and follow-up, t(35) = 1.44, p = .16.

Discussion

The present study used referential cohesion and semantic cohesion to select written materials written with high and low text cohesion. Specifically, CRC prevention information that is high in text cohesion contains higher rates of argument overlaps and greater conceptual and semantic similarity between sentences. Our study supports the first hypothesis that CRC prevention information written with high text cohesion would be easier to read by showing faster reading times. After adjusting for the tradeoff between reading times and text comprehension, our study supports the second hypothesis that CRC prevention information written with high text cohesion would be easier to understand. Our third hypothesis was not supported; information with high text cohesion was not shown to be easier to remember as measured by word recognition and recall scores.

Our findings agree with prior research that has shown shorter reading times for high-cohesion text compared to low-cohesion text (McNamara & Kintsch, 1996). Text cohesion helps readers to connect ideas and derive meaning from text without slowing down the vertical and cyclic process of reading (Kintsch, 1998). Specifically, highly cohesive text is read faster because readers do not need to stop to retrieve earlier ideas that have been stored in the working memory or stop to make an inference. On the other hand, text with low cohesion requires readers to make inferences based on background knowledge to bridge cohesion gaps (Singer & Ritchot, 1996). Participants in this study did not have extensive knowledge of colorectal cancer and screening tests before the study. Since they did not have sufficient background knowledge to draw inferences while reading low cohesive text, they needed to slow down reading to draw inferences from earlier text stored in working memory.

Prior researchers have tried to improve text cohesion by increasing argument overlaps and writing out implicit information based on the contemporary comprehension theory (Britton & Gu˝lgo˝z, 1991). These researchers found that comprehension and immediate recall were higher for the revised text than they were for the original text. A later study was able to successfully replicate these findings (McNamara & Kintsch, 1996). Our initial analyses do not show differences between high text cohesion and low text cohesion in text comprehension and recall scores. Researchers have suggested the need to take reading times into account when measuring comprehension because the challenge of reading difficult text can be manifested through either prolonged reading times or decreased comprehension scores (Kintsch & van Dijk 1978). Our post hoc analyses yield some evidence to support the effect of text cohesion on comprehension. A plausible interpretation of this finding is that participants allocated more time to read low cohesive text than high cohesive text in order to achieve the same level of text comprehension.

Moreover, the present study identified a strong effect of subtopic on reading times and word recognition scores. Participants read much faster and recognized more words in the last two subtopics than the first two subtopics. A higher level of knowledge can reduce reading times and increase reading efficiency in older readers (Miller, 2009; Miller & Stine-Morrow, 1998). Participants might have had more knowledge about risk factors for CRC and screening tests than about locations of the colon and rectum and also symptoms of CRC. Nevertheless, their scores are only modest on the test of prior knowledge in which over half of test questions are related to CRC risk and screening tests. Alternatively, the recency effect may account for the subtopic effect. The recency effect means that people have superior memory of words at the end of a list to which they were mostly recently exposed (Deese & Kaufman, 1957). Another plausible explanation could be that text cohesion level, specifically referential cohesion, in subtopics 3 and 4 is actually relatively higher than it is in subtopics 1 and 2 in both experimental texts. Higher rates of argument overlaps in subtopics 3 and 4 could have enhanced reading speed and word recognition.

According to contemporary comprehension theory (Kintsch, 1998), text cohesion helps readers connect ideas to build a cohesive representation of the text. The links among ideas are stronger in text written with high cohesion than in text written with low cohesion. Consequently, information written with high-cohesion text is anticipated to be stored longer in memory than low-cohesion text. However, neither word recognition nor recall showed significant results.

This study has several limitations hat should be considered when interpreting results. First, the sample size is small because of the pilot nature of this study. Second, more than half of the sample was adherent to CRC screening guidelines. Third, the majority of study participants have adequate health literacy. Although the short version of the Test of Functional Health Literacy in Adults is a well-recognized health literacy test, it may not be the best measure to assess participants' health literacy skills beyond word level because of its Cloze test format. Furthermore, the cross-over study design tangled subtopic effect with text cohesion effect. A between-subject design may be more suitable to examine the effect of text cohesion and avoid other confounding text factors. Last, the present study did not focus on global cohesion which is the logical organization of ideas in a text (Albrecht & O'Brien, 1993). A recent study has shown that improving global cohesion by grouping related information into sections increases people's recall of a clinic visit note (Smith, Hetzel, Dalrymple, & Keselman, 2011).

The present study reveals some benefits to writing CRC prevention information with high text cohesion. CRC prevention information written at higher text cohesion facilitates reading speed and comprehension, but not retention. Lowering reading grade level has long been recommended to improve readability of health information. Alternatively, the present study suggests that increasing text cohesion—referential cohesion and semantic cohesion—may also be beneficial. Referential cohesion can be increased by repeating key ideas and semantic information can be increased by adding relevant knowledge to the information.

This study sheds some light on the value of increasing text cohesion to enhance readability of written CRC prevention information, which has potential to increase people's health literacy about cancer screening; that is, to increase their ability to process and understand basic cancer prevention information needed to get screened. People often seek cancer prevention information from print or online written materials (Ling, Klein, & Dang, 2006). Text cohesion can reduce reading times without compromising one's the ability to understand written information. Further research on the effects of text cohesion on comprehension and retention of cancer prevention and other written health information is warranted.

Acknowledgments

This study was supported by the Research Support Funds Grant from Indiana University-Purdue University Indianapolis to the first author.

Notes

^a Average scores of digit forward and digit backward.

*p < .05. **p < .001.

Appendix: General Knowledge of Colon Cancer and Screening Questionnaire

Here are some questions about colon cancer. I'll read the questions and the answer choices. You tell me what you think is the best answer. Or, you can answer “I don't know.”

1.	Can colon cancer ever be prevented? Would you say □ 1. Yes □ 2. No, or □ 8. Don't know
2.	Who is more likely to get colon cancer? Would you say □ 1. A person younger than 50 years old □ 2. A person older than 50 years old □ 3. There's no difference, or □ 8. Don't know
3.	Who is more likely to get colon cancer? Would you say □ 1. Someone whose husband or wife had colon cancer □ 2. Someone with one close blood relative, like a parent, brother or sister, who had colon cancer □ 3. Someone with two close blood relatives, like a parent, brother, or sister, who had colon cancer □ 4. There's no difference, or □ 8. Don't know
4.	Is a woman's chance of getting colon cancer… □ 1. Much higher than a man's? □ 2. About the same as a man's? □ 3. Much lower than a man's?, or □ 8. Don't know
5.	Which of the following descriptions can be a symptom of colon cancer? □ 1. Having diarrhea or constipation □ 2. Finding your stools are narrower than usual □ 3. 1 and 2, or □ 8. Don't know
6.	Which of these is the most effective way for people to lower their chances of dying from colon cancer? Would you say □ 1. Exercising regularly □ 2. Finding and removing polyps □ 3. Limiting alcohol □ 4. There's nothing people can do to lower their chance of dying from colon cancer, or □ 8. Don't know
7.	What can be found by doing a stool blood test, which might also be called a Hemoccult or fecal occult blood test? □ 1. Protein in your blood □ 2. How much fiber you have been eating □ 3. Hidden blood in your stool, or □ 8. Don't know
8.	If you choose to have a stool blood test and everything is normal, when will you need to have your next one? Would you say □1. In six months □2. In one year □3. In two years □4. In five years □5. In ten years, or □8. Don't know
9.	What is a doctor able to see during a colonoscopy? Would you say □ 1. Inside only the lower part of your colon □ 2. Hidden blood in your stool □ 3. Inside the entire length of your colon, or □ 8. Don't know
10.	If you choose to have a colonoscopy and everything is normal, when will you probably need to have your next one? Would you say □ 1. In six months □ 2. In one year □ 3. In two years □ 4. In five years □ 5. In ten years, or □ 8. Don't know