Final Project- Cancer Survivorship and Disability

1 Introduction

1.1 Background

There has been a growing population of cancer survivors because of recent advancements in cancer treatment. Despite these advancements, more knowledge is still needed in the field of cancer survivorship. According to the American Cancer Society, the 5-year relative survival rate for all cancers combined is 68% as of 2017. Beneath the success seen with such statistic is the emerging and often overlooked public health problem of disability after cancer treatment. According to Batai et al. (2022), 40% of cancer survivors experience long-term effects/disability from cancer. Understanding the relationship between cancer survivorship and disability can provide insight into providing care for cancer survivors and advancing knowledge in the survivorship phase of cancer care.

1.2 Question of interest

What is the relationship between cancer survivorship and disability?

• Secondary question: And how do the variables of smoking, age, and chronic health condition contribute to the odds of disability?

Variables of interest:

• Outcome variables: Visual disability, hearing disability, mobility disability, and disability in activities in daily life (ADL)*
• Primary predictor variable: Cancer survivorship
• Possible confounders: Healthcare coverage, smoking, and age
• Potential effect modifiers: Chronic health condition

*NY BRFSS defined visual disability as being blind or having difficulty seeing, hearing disability as being deaf or having difficulty hearing, mobility disability as difficulty walking, and disability in ADL as difficulty with activities such as dressing, bathing, and doing errands alone.

1.3 Data set

For this report, I used data from the 2020 New York State Behavioral Risk Factor Surveillance System (NY BRFSS), which is modeled on the National Behavioral Risk Factor Surveillance System (BRFSS). BRFSS is a national health survey system conducted by the Centers for Disease Control and Prevention that aims to collect self-reported state data on U.S. adult residents’ modifiable risk behaviors, health conditions, and use of preventive care, leading the unit of observation to be individual U.S. adults. The survey is administered through telephone and data is collected yearly. It includes demographic, health behavior, and health condition questions. As stated above, the 2020 NY BRFSS is the specific data set that I used in this report, which has a sample of 14,769 individuals that is representative of the non-institutionalized civilian adult population (18 years or older) in the State of New York.

2 Exploratory Data Analysis

2.1 Exploring the relationship of interest

Firstly, I studied the question of interest What is the relationship between cancer survivorship and disability? through an exploratory data display.

Figure 2.1: Bar graphs for risk (%) of disabilities related to hearing, visual, mobility, and activities of daily living (ADL), comparing cancer survivor status

Discussion of Fig. 2.1: Of the 14,769 adults who participated and provided data for the 2020 NY BRFSS, risk of various disabilities varied between those who reported having had cancer (cancer survivor) and those who reported not having had cancer (non-survivor). Overall, cancer survivors had a higher risk of disability compared to non-survivors for all four disabilities. The risk of hearing disability (11.8%) among cancer survivors was higher than the risk hearing disability (5.2%) among non-survivors. A similar trend was seen with mobility, as the risk of mobility disability (22.5%) among cancer survivors was higher than the risk of mobility disability (12.1%) among non-survivors. For visual and activities in daily life (ADL), the difference in risk was relatively smaller. The risk of visual disability (5.6%) among cancer survivors was slightly higher than the risk of visual disability (4.1%) among non-survivors. The risk of disability in ADL (11.1%) among cancer survivors was slightly higher than the risk of disability in ADL (7%) among non-survivors.

2.2 Exploring confounders

For the following section, I explored potential confounders of the relationship between cancer survivorship and disability by answering the question: Are there other variables that explain the association between (and are related to) the predictor and outcome variables? As confounders are variables associated with both the predictor and outcome, I created data displays to verify if healthcare coverage, smoking status, and age are confounders and thus should be included in the finalized model.

2.2.1 Comparing healthcare coverage

2.2.1.1 Risk (%) of cancer survivorship

Figure 2.2: Bar graph for risk (%) of cancer survivorship comparing healthcare coverage

2.2.1.2 Risk (%) of disability

Figure 2.3: Bar graph for risk (%) of disability comparing healthcare coverage

Fig. 2.2 showed that the risk of cancer survivorship (14%) (ie. being a cancer survivor) among individuals with healthcare coverage was higher than the risk of cancer survivorship (4.6%) among individuals without healthcare coverage, showing that healthcare coverage seemed to be positively associated with cancer survivorship.

Fig. 2.3 showed that the risk of hearing disability (5.9%) and mobility disability (12.8%) among individuals with healthcare coverage were respectively higher than the risk of hearing disability (4.2%) and mobility disability (11.5%) among individuals without healthcare coverage. However, such positive association was very minimal for disability in ADL, as the risk of disability in ADL was 7.1% among individuals with healthcare coverage and 6.8% among individuals without healthcare coverage. Moreover, there was a negative association for visual disability, as risk of visual disability (4.1%) among individuals with healthcare coverage was lower than the risk of visual disability (4.5%) among individuals without healthcare coverage. Therefore, although there seemed to be an association between healthcare coverage and cancer survivorship, there did not seem to be an association between healthcare coverage and disability, leading to its exclusion in the finalized model.

2.2.2 Comparing smoking status

2.2.2.1 Risk (%) of cancer surivorship

Figure 2.4: Bar graph for risk (%) of cancer survivorship comparing smoking status

2.2.2.2 Risk (%) of disability

Figure 2.5: Bar graph for risk (%) of disability comparing smoking status

Fig. 2.4 showed that the risk of cancer survivorship (18%) among smokers was higher than the risk of cancer survivorship (10.5%) among non-smokers, showing that smoking status seemed to be positively associated with cancer survivorship.

Fig. 2.5 showed that for all four disabilities, the risk of disability among smokers was higher than the risk of disability among non-smokers. The highest risk of a disability was among smokers: the risk of mobility disability being 18.9% among smokers. The lowest risk of a disability was among non-smokers: the risk of visual disability being 3.4% among non-smokers.

Therefore, Fig. 2.4 and Fig. 2.5 ensured the appropriateness to include smoking status when selecting a final model.

2.2.3 Comparing age groups

2.2.3.1 Risk (%) of cancer surivorship

Figure 2.6: Bar graph for risk (%) of cancer survivorship comparing age groups

2.2.3.2 Risk (%) of disability

Figure 2.7: Bar graph for risk (%) of disability comparing age groups

Fig. 2.6 showed that the overall risk of cancer survivorship increased as the age group increased, with the highest risk of cancer survivorship (27.9%) seen among individuals 65 years or older.

Fig. 2.7 showed that the overall risk of each of the four disabilities increased as the age group increased. For example, the risk of hearing disability was 0.8% in the [18, 24] age group, 1.3% in the [25, 34] age group, 1.9% in the [35, 44] age group, 2.6% in the [45, 54] age group, 4.9% in the [55, 64] age group, and 12.8% in the 65 or older age group. A similar trend can be seen for the other disabilities. The only difference in this trend is that the risk of disability in ADL decreased from 3.4% in the [18, 24] age group to 2.7% in the [25, 34] age group. However, this deviation from the trend was minimal and the overall trend seemed to be that age is positively associated with disability.

Therefore, Fig. 2.6 and Fig. 2.7 ensured the appropriateness to include age when selecting a final model.

2.3 Exploring effect modification

In the following section, I explored chronic health condition as a potential effect modifier of the relationship between cancer survivorship and disability by answering the question: Does the contribution of survivorship to the risk of disability vary by chronic health condition?

Figure 2.8: Bar graphs for relative risk (RR) of various disabilities (comparing cancer survivorship status) for individuals with chronic health conditions compared to those with no chronic health condition

Fig. 2.8 showed that the overall trend was that the relative risk (RR) of disability for cancer survivors compared to non-cancer survivors was larger for individuals with chronic health condition than those without chronic health condition. The largest increase in RR can be seen with disability in ADL: the risk of disability in ADL for a cancer survivor with chronic health condition compared to non-cancer survivors with chronic health condition (RR=1.9) was larger than the risk of disability in ADL for a cancer survivor without chronic health condition to a non-cancer survivor without chronic health condition (RR=0.71), a difference of 1.19. Therefore, survivor*chronic health condition was initially included as an interaction term when selecting a final model.

3 Regression Analysis

3.1 Analysis plan

Since there were four outcome variables (ie. visual disability, hearing disability, mobility disability, and disability in ADL), I planned to create four models with the same predictor and confounding variables for each outcome variable. Moreover, the outcome variables of interest were binary, so logistic regression was used as the type of analysis. To control for confounding, confounding variables were included in the models.

3.2 Final model selection procedure

Figures 2.2-2.7 provided insight to my initial curiosity of the confounding and effect modifying variables. Based on these data displays, I excluded healthcare coverage from the final models because Fig. 2.3 showed no association between healthcare coverage and disability. Based on an initial logistic regression analysis, the coefficients associated with the interaction term survivor*chronic health condition were not statistically different from 1; and thus, was also excluded in the final models. However, chronic health condition as a confounder was statistically different from 1 for all outcome variables/models, so I included it as a confounder variable in the final models. Further logistic regression analysis also showed that the coefficients associated with the smoker variable were not statistically significant in Model 1 (95% CI for OR: 0.99, 1.57), but they were statistically significant in Model 2 (95% CI for OR: 1.03, 1.70), Model 3 (95% CI for OR: 1.08, 1.49), and Model 4 (95% CI for OR: 1.07, 1.61). Similarly, the coefficients associated with the age groups were not always statistically significant in Models 1, 2, and 4. Thus, I used Akaike’s “An Information Criterion” (AIC) approach to see which models (the ones including all the variables or the ones excluding statistically insignificant variables) fit the data best. For the outcome of hearing disability, Model 1 (AIC = 4256.7) was the final model as it had a lower AIC than Model 1a (AIC = 4880.5). For the outcome of visual disability, Model 2 (AIC = 4286.1) was the final model as it had a lower AIC than Model 2a (AIC = 4354.2). For the outcome of disability in ADL, Model 4 (AIC = 6109.6) was the final model as it had a lower AIC than Model 4a (AIC = 6290.6). In summary, I fitted a logistic regression including an indicator of cancer survivorship, smoking status, age (in 6 categories), and an indicator of chronic health condition.

3.3 Comparison of model outputs

In the concluding section of my report, I aimed to answer the primary research question of interest What is the relationship between cancer survivorship and disability? and the secondary question And how do the variables of smoking, age, and chronic health condition contribute to the odds of disability? by comparing the output of all four models.

(\#tab: )Comparison of four model outputs for odds of disability for smoking status, age, and chronic health condition

Characteristic	Hearing disability (Model 1)			Visual disability (Model 2)			Mobility disability (Model 3)			Disability in ADL (Model 4)
	OR1	95% CI1	p-value	OR1	95% CI1	p-value	OR1	95% CI1	p-value	OR1	95% CI1	p-value
Cancer Survivor
No	—	—		—	—		—	—		—	—
Yes	1.07	0.81, 1.40	0.6	1.08	0.75, 1.56	0.7	0.98	0.78, 1.22	0.8	1.37	1.00, 1.87	0.049
Smoker
No	—	—		—	—		—	—		—	—
Yes	1.25	0.99, 1.57	0.064	1.32	1.03, 1.70	0.028	1.27	1.08, 1.49	0.004	1.31	1.07, 1.61	0.010
Age
18 - 24	—	—		—	—		—	—		—	—
25 - 34	1.31	0.43, 4.06	0.6	1.13	0.55, 2.32	0.7	1.24	0.56, 2.74	0.6	0.63	0.33, 1.21	0.2
35 - 44	2.15	0.73, 6.33	0.2	1.37	0.68, 2.75	0.4	4.02	1.93, 8.38	<0.001	1.31	0.75, 2.29	0.4
45 - 54	2.44	0.84, 7.10	0.10	1.58	0.80, 3.11	0.2	5.41	2.64, 11.1	<0.001	1.23	0.72, 2.10	0.4
55 - 64	5.74	2.05, 16.1	<0.001	1.51	0.81, 2.83	0.2	8.04	3.98, 16.2	<0.001	1.62	0.96, 2.72	0.069
65 or older	14.3	5.21, 39.0	<0.001	1.96	1.05, 3.67	0.035	12.5	6.21, 25.2	<0.001	1.91	1.14, 3.21	0.015
Chronic Health Condition
No	—	—		—	—		—	—		—	—
Yes	2.08	1.61, 2.70	<0.001	2.97	2.26, 3.91	<0.001	5.00	4.19, 5.97	<0.001	4.12	3.27, 5.18	<0.001
1 OR = Odds Ratio, CI = Confidence Interval

Discussion of Table 1:

Table 1 showed a comparison of outputs between the four models (that only differ by outcome variable). In Model 1, the odds of hearing disability among cancer survivors was 1.07 times the odds of hearing disability among otherwise similar non-cancer survivors, as smoker, age, and chronic health condition were being held constant. However, the 95% CI for the OR was (0.81,1.40); and thus, not statistically significant. Holding all other variables constant, the odds of hearing disability among smokers was 1.25 times the odds of hearing disability among non-smokers. However, this difference is not statistically significant (95% CI for OR: 0.99, 1.57). Only the ORs of the 55 to 64 and 65 or older age groups were statistically significant. Holding all other variables constant, individuals who were 55 to 64 years old had a 474% increased odds of hearing disability compared to those who were 18 to 24 years old (95% CI for OR: 2.05, 16.1). Holding all other variables constant, individuals who were 65 years old or older had a 1330% increased odds of hearing disability compared to those who were 18 to 24 years old (95% CI for OR: 5.21, 39.0). Individuals with chronic health condition had a 108% increased odds of hearing disability compared to otherwise similar individuals without chronic health condition, as survivor status, smoker, and age were being held constant (95% CI for OR: 1.61, 2.70).

In Model 2, the odds of visual disability among cancer survivors was 1.08 times the odds of visual disability among otherwise similar non-cancer survivors, as smoker, age, and chronic health condition were being held constant. However, the 95% CI for the OR was (0.75, 1.56); and thus, not statistically significant. Holding all other variables constant, smokers had a 32% increased odds of visual disability compared to non-smokers (95% CI for OR: 1.03, 1.70). Holding all other variables constant, individuals who were 65 years of older had a 96% increased odds of visual disability compared to those who were 18 to 24 years old (95% CI for OR: 1.05, 3.67), which was the only age group with an OR that was statistically significant. Individuals with chronic health condition had a 197% increased odds of visual disability compared to otherwise similar individuals without chronic health condition, as survivor status, smoker, and age were being held constant (95% CI for OR: 2.26, 3.91).

In Model 3, the odds of mobility disability among cancer survivors was 0.98 times the odds of mobility disability among otherwise similar non-cancer survivors, as smoker, age, and chronic health condition were being held constant. However, the 95% CI for the OR was (0.78, 1.22); and thus, not statistically significant. Holding all other variables constant, smokers had a 27% increased odds of mobility disability compared to non-smokers (95% CI for OR: 1.08, 1.49). Only the 25 to 34 years old age group had an OR that was not statistically significant (95% CI for OR: 0.56, 2.74). Compared to otherwise similar individuals who were 18 to 24 years old, those who were 35 to 44 years had a 302% increased odds of mobility disability (p < 0.001), those who were 45 to 54 years had a 441% increased odds of mobility disability (p < 0.001), those who were 55 to 64 years had a 704% increased odds of mobility disability (p < 0.001), and those who were 65 years or older had a 1150% increased odds of mobility disability (p < 0.001). Individuals with chronic health condition had a 400% increased odds of mobility disability compared to otherwise similar individuals without chronic health condition, as survivor status, smoker, and age were being held constant (95% CI for OR: 4.19, 5.97).

In Model 4, cancer survivors had a 37% increased odds of disability in ADL compared to otherwise similar non-cancer survivors, as smoker, age, and chronic health condition were being held constant. Unlike other disabilities/models, this OR in Model 4 was statistically significant, as the 95% CI for the OR was (1.00, 1.87). Holding all other variables constant, smokers had a 31% increased odds of disability in ADL compared to non-smokers (95% CI for OR: 1.07, 1.61). Holding all other variables constant, individuals who were 65 years or older had a 91% increased odds of disability in ADL compared to those who were 18 to 24 years old (95% CI for OR: 1.14, 3.21), which was the only age group with an OR that was statistically significant. Individuals with chronic health condition had a 312% increased odds of disability in ADL compared to otherwise similar individuals without chronic health condition, as survivor status, smoker, and age were being held constant (95% CI for OR: 3.27, 5.18).