Retention for the 490 initially enrolled participants (Figure 1) was high (69% of participants completed the 24-month assessment) and did not differ significantly between survivors and controls (p=0.34). Specifically, 84% of participants completed at least 2 assessments, 73% completed 3, and 59% completed all 4. Completion of the 24-month assessment did not vary significantly by participant age or clinical site but did differ by WRAT (p=0.01) and Time 1 domain scores (all p<0.01); those who completed had slightly higher WRAT and domain scores.
Table 1 provides a description of the sample. Survivors and controls were well matched on education, but survivors not treated with chemotherapy were significantly older than survivors treated with chemotherapy and controls. The majority of participants were retired (72%) or unemployed (5%) and most were married (70%).
Table 1
Participant characteristics by treatment group
|
No. of Patients (%)
|
|
Characteristic
|
Overall
|
Chemo
|
No Chemo
|
Control
|
p-value
|
|
(n = 490)
|
(n = 160)
|
(n = 168)
|
(n = 162)
|
|
Age, M (SD) [n = 489]
|
72.6 (6.0)
|
70.9 (5.1)
|
74.6 (5.9)
|
72.1 (6.4)
|
<.001
|
Race
|
|
|
|
|
|
White
|
415 (85%)
|
129 (81%)
|
141 (84%)
|
145 (90%)
|
.282
|
Black
|
37 (8%)
|
16 (10%)
|
15 (9%)
|
6 (4%)
|
|
Asian/ PI
|
17 (3%)
|
8 (5%)
|
5 (3%)
|
4 (2%)
|
|
Other
|
14 (3%)
|
5 (3%)
|
5 (3%)
|
4 (2%)
|
|
Missing
|
7 (1%)
|
2 (1%)
|
2 (1%)
|
3 (2%)
|
|
Ethnicity
|
|
|
|
|
|
Hispanic
|
49 (10%)
|
18 (11%)
|
12 (7%)
|
19 (12%)
|
.310
|
Non-Hispanic
|
432 (88%)
|
138 (86%)
|
153 (91%)
|
141 (87%)
|
|
Missing
|
9 (2%)
|
4 (3%)
|
3 (2%)
|
2 (1%)
|
|
Education
|
|
|
|
|
|
Less than college
|
196 (40%)
|
67 (42%)
|
71 (42%)
|
58 (36%)
|
.382
|
College or more
|
292 (60%)
|
92 (58%)
|
96 (57%)
|
104 (64%)
|
|
Missing
|
2 (0%)
|
1 (1%)
|
1 (1%)
|
0 (0%)
|
|
Smoking Hx
|
|
|
|
|
|
Yes
|
224 (46%)
|
85 (53%)
|
74 (44%)
|
65 (40%)
|
.049
|
No
|
264 (54%)
|
74 (46%)
|
93 (55%)
|
97 (60%)
|
|
Missing
|
2 (0%)
|
1 (1%)
|
1 (1%)
|
0 (0%)
|
|
Endocrine Therapy
|
|
|
|
|
|
Ever (n = 314)
|
234 (75%)
|
110 (72%)
|
124 (77%)
|
NA
|
.298
|
At Assessment 1 (n = 319)
|
80 (25%)
|
52 (34%)
|
28 (17%)
|
NA
|
<.001
|
Cancer Characteristics
|
|
|
|
|
|
ER Positive (n = 297)
|
237 (80%)
|
111 (73%)
|
126 (88%)
|
NA
|
.001
|
PR Positive (n = 292)
|
190 (65%)
|
84 (55%)
|
106 (76%)
|
NA
|
<.001
|
HER2 (FISH) Positive (n = 266)
|
32 (12%)
|
25 (17%)
|
7 (6%)
|
NA
|
.006
|
Tumor size (cm), M (SD)
|
1.7 (1.4)
|
2.2 (1.5)
|
1.2 (1.2)
|
NA
|
<.001
|
Years since diagnosis, M (SD)
|
8.0 (2.7)
|
8.1 (2.7)
|
8.0 (2.6)
|
NA
|
.574
|
Baseline Psych [n = 488]
|
|
|
|
|
|
FSI Disruption, M (SD)
|
8.1 (11.0)
|
9.5 (12.4)
|
8.9 (10.8)
|
5.9 (9.3)
|
.008
|
STAI State Sum, M (SD)
|
25.8 (7.3)
|
26.8 (8.3)
|
25.5 (6.9)
|
25.0 (6.6)
|
.076
|
CESD Sum, M (SD)
|
6.8 (7.1)
|
7.9 (8.8)
|
6.6 (6.1)
|
5.9 (5.9)
|
.043
|
Genotyping was available for 472 participants. Over one fifth (22%) of the participants were ε4 carriers (23% survivors vs. 18% controls, p=0.21); of these, only 4 were homozygous for ε4. Overall, 46% of participants had a history of smoking (49% survivors vs. 40% controls, p=0.09). Of those with a smoking history, 95% started by the age of 25, with median start age of 17 years. Almost all (85%) of the participants with a smoking history reported quitting more than 10 years prior to study enrollment.
Survivors had significantly lower scores on all domains, even after adjusting for practice effects and covariates (i.e., age, CESD, FSI, WRAT). These group differences persisted over time. Observed deficits in survivors collapsed over assessment time points, as compared to controls, were -0.22, -0.15, and -0.19 for Language, Attention, Processing Speed, Executive Function (APE), and Learning and Memory domains, respectively. Median adjusted estimates for deficits in survivors, as compared to controls, were -0.20, -0.14, and -0.14 points, respectively. However, no significant differences were seen when comparing survivors treated with chemotherapy vs. those not exposed to chemotherapy.
Figure 2 plots the observed average domain scores over time (left panel). A small practice effect due to repeated test administration was observed. At each assessment, domain scores increased, on average, 0.04, 0.02, and 0.14, for Language, APE, and Learning and Memory, respectively. This practice effect did not differ significantly between survivors and controls for APE or Learning and Memory domains, but differed significantly for the Language domain, with healthy controls exhibiting a slightly stronger practice effect compared to controls (p=0.01).
Also plotted in Figure 2, right panel, are the model-estimated median domain scores, stratified by ε4 status and cohorts, collapsed over assessment time points to facilitate the visual inspection of the group differences. As shown in the figure, the ε4+survivor cohort had the lowest average domain scores compared to ε4-survivors or healthy controls of either ε4 status. The overall pattern of the model-estimated medians suggested the possibility of interactions between APOEε4 status and cohorts to address the first two research questions outlined in the Introduction. Thus, the domain scores were evaluated in a model on survivorship, APOEε4 status, the survivorship and APOEε4 interaction, and controlling for time and covariates, as described above. The APOEε4 status by cohort interaction term addressed whether or not APOEε4 was associated with a greater cognitive impact in cancer survivors than controls. It was evaluated by a contrast between ε4 carriers and ε4 non-carriers across the survivors and controls. The bar plot on the Learning and Memory domain shows that ε4+ survivors had a visibly lower score than ε4- survivors. Specifically, ε4- survivors had a median score of 0.09, whereas ε4+ survivors had a considerably worse median score at –0.11. This translates to a contrast of –0.11 – 0.09 = –0.20 (i.e., ε4+ survivors fared worse than ε4- survivors). By comparison, the difference was smaller in the non-cancer control cohort, where the ε4- controls had a median score of 0.23 and ε4+ controls had a median score of 0.18,, which represented a difference of 0.18 – 0.23 = –0.05. Thus, the two-way interaction involved the scaled contrast of these two differences, which was (–0.20 – (–0.05))/4 = –0.04. The corresponding 95% Bayesian HDI was between (-0.06, -0.01), indicating that as compared to the controls, ε4+ survivors fared worse than ε4-survivors in the Learning and Memory domain, and the posterior confidence on this difference was above 95%. Next, we examined the APE domain (middle bar plot), where a similar detrimental effect was also found in ε4+survivors (median contrast = -0.04, 95% HDR: -0.05, -0.03). The language domain (top) had a stronger effect size (median contrast = -0.05, 95% HDI: -0.07, -0.03), as the ε4 effect was reversed amongst controls. Within both the APE and Learning and Memory domains, although there were large main effects associated with both APOE status and survivorship, the ε4 effect was stronger among survivors even after adjustment for covariates.
We next assessed the 3-way interaction between APOE status, survivorship, and smoking history with adjustment for covariates. This 3-way interaction addressed the additional research question that smoking confers a differential protective effect for ε4+ survivors, who in the analyses above showed a greater vulnerability than ε4- survivors. For instance, in the Language domain, the 3-way interaction contrast had a median of 0.20 (95% HDR: 0.15, 0.24). This contrast effectively represents that the difference between ε4+survivors with a smoking history compared to ε4+ survivors without a smoking history may be significantly different than the difference between ε4+ control with a smoking history compared to ε4+ controls without a smoking history, and this difference of differences was greater than zero above a 95% posterior confidence. Figure 3 provides a visualization of this complex 3-way interaction. Similarly, a statistically reliable effect was found in the APE domain (APE median contrast = 0.10, 95% HDR: 0.07, 0.13) and Learning & Memory (median contrast = 0.10, 95% HDI: 0.04, 0.15), although these effect sizes were only half of that in the Language domain. Note that ε4+ smokers scored visibly greater than ε4+ non-smokers, across survivorship status and domain.
Specific interpretation of the 3-way interaction effects is best assessed via 2-way interactions stratified by survivorship; the compensatory effect of smoking on the effect of APOE ε4 carriage was largely driven by controls. For example, the median estimate for the 2-way interaction for the APE domain among survivors was 0.04 (95% HDI = -0.25, –0.34), while the median estimate among controls was 0.48 (95% HDI = 0.04, 0.92). Specific contrasts between smokers and non-smokers of each ε4+ group are given in Table 2. For controls, smoking had a protective effect on ε4-status nearly twice that of survivors (median for controls: 0.43 vs. median for survivors: 0.23). That is, smoking protected both survivors and controls, but the protective effect was stronger for controls. The difference was even more pronounced for Learning and Memory (median for controls: 0.36 vs. median for survivors: 0.04). Mean scores by APOE ε-status, survivorship, and smoking history, based on the Bayesian HLM models controlling for smoking, WRAT, CESD, and FSI, are depicted in Figure 3.
Table 2
Neuropsychological and Self-Report Measures
Cognitive Reserve
|
Wide Range Achievement Test 4 (WRAT4 Reading) (31)
|
Language
|
Category Fluency (32); Boston Naming Test (33), Controlled Oral Word Association Test (34)
|
Attention, Processing Speed, Executive Function
|
Digit Symbol (33); Trail Making A and B (35); DKEFS Color-Word Naming (36); NAB Digits Forward and Backward (37-38); NAB Driving Scenes (37-28).
|
Learning and Memory
|
NAB List Learning (37-38) Trial 1, Semantic Clustering, List A Immediate, List A Delayed, Long Delay, List B Immediate, New Recognition Index; Logical Memory Part 1 and Story B (WMS-R, 39).
|
Self-Report Questionnaires
|
Center for Epidemiological Study – Depression, (40); Spielberger State Anxiety Inventory, (41); Fatigue Symptom Inventory, 42); Smoking history: CDC-Behavioral Risk factor Surveillance System (43).
|