Association between diet quality and ovarian cancer risk and survival

Abstract

Background

Research on diet quality and ovarian cancer is limited. We examined the association between diet quality and ovarian cancer risk and survival in a large prospective cohort.

Methods

We used data from women in the prospective National Institutes of Health–AARP Diet and Health Study enrolled from 1995 to 1996 who were aged 50-71 years at baseline with follow-up through December 31, 2017. Participants completed a 124-item food frequency questionnaire at baseline, and diet quality was assessed via the Healthy Eating Index-2015, the alternate Mediterranean diet score, and the Dietary Approaches to Stop Hypertension score. Primary outcomes were first primary epithelial ovarian cancer diagnosis from cancer registry data and among those diagnosed with ovarian cancer all-cause mortality. We used a semi-Markov multistate model with Cox proportional hazards regression to account for semicompeting events.

Results

Among 150 643 participants with a median follow-up time of 20.5 years, 1107 individuals were diagnosed with a first primary epithelial ovarian cancer. There was no evidence of an association between diet quality and ovarian cancer risk. Among those diagnosed with epithelial ovarian cancer, 893 deaths occurred with a median survival of 2.5 years. Better prediagnosis diet quality, according to the Healthy Eating Index-2015 (quintile 5 vs quintile 1: hazard ratio [HR] = 0.75, 95% confidence interval [CI] = 0.60 to 0.93) and alternate Mediterranean diet score (quintile 5 vs quintile 1: HR = 0.68, 95% CI = 0.53 to 0.87), was associated with lower all-cause mortality. There was no evidence of an association between Dietary Approaches to Stop Hypertension score and all-cause mortality.

Conclusions

Better prediagnosis diet quality was associated with lower all-cause mortality after ovarian cancer diagnosis but was not associated with ovarian cancer risk.

With 19 680 ovarian cancer patients (90% of which are epithelial carcinomas) and 12 740 deaths projected in the United States in 2024 (1), ovarian cancer is the leading cause of gynecological cancer–related deaths. The 5-year survival rates for stage III and stage IV patients are 42% and 26%, respectively. With 60% of ovarian cancer patients diagnosed at these late stages (2), this cancer is highly relevant for exploring modifiable factors for risk and survival (1).

The World Cancer Research Fund–American Institute for Cancer Research’s latest summary of the association between diet and ovarian cancer risk was “limited-no conclusion” because of limited studies (3). Similarly, though commonly used for other cancers (4,5), few studies of ovarian cancer have evaluated score-based measures of diet quality (6-13), a comprehensive approach accounting for the biological interdependence of individual nutrients in multiple dietary components (14). Poor diet encompassing pro-inflammatory foods, such as refined carbohydrates, red and processed meat, and fried food, may promote alterations in sex steroid (15) or peptide hormones (eg, elevated circulating gonadotropins) (16), glucose homeostasis (17) and inflammatory cytokines (18,19), which may contribute to ovarian carcinogenesis. Similarly, overall diet quality could impact ovarian cancer outcomes through metabolic pathways and immune and/or inflammatory effects directly or through adiposity (20) or interact with cancer treatment through chemotherapy drug pharmacokinetics (20).

Two case-control studies (6,7) found better diet quality was associated with lower ovarian cancer risk, whereas earlier cohort (8,9) and case-control (10) studies were null. Research on diet quality and ovarian cancer survival is more limited (11-13). Ovarian cancer survivors in the Women’s Health Initiative with higher prediagnosis diet quality per the Healthy Eating Index-2005 (HEI-2005) had 27% lower all-cause mortality (11). In a Chinese cohort, higher postdiagnosis Chinese HEI scores were associated with better overall survival (hazard ratio [HR] = 0.56, 95% confidence interval [CI] = 0.36 to 0.88) (13). However, a recent analysis in the prospective Ovarian Cancer Prognosis and Lifestyle (OPAL) study (12) found no association for pre- or postdiagnosis diet quality via 4 dietary scores with ovarian cancer survival.

Because of the small number of ovarian cancer patients in earlier cohort studies, the inherent limitations of assessing diet in cancer case-control studies, and the lack of modeling competing events, large prospective studies of diet quality and ovarian cancer risk and survival are needed to inform dietary recommendations and lifestyle interventions for ovarian cancer prevention and survivorship. Therefore, we examined the associations between prediagnosis diet quality and ovarian cancer risk and survival in the large, prospective National Institutes of Health–AARP (NIH-AARP) Diet and Health Study (21).

Methods

Study population

The prospective NIH-AARP Diet and Health Study enrolled 566 398 participants between 1995 and 1996 (21) who were members of the AARP (formally the American Association of Retired Persons), aged 50-71 years at baseline residing in 6 states (California, Florida, Louisiana, New Jersey, North Carolina, Pennsylvania) or 2 metropolitan areas (Atlanta, Georgia; Detroit, Michigan). Participants completed baseline questionnaires on demographic, dietary, medical, anthropometric, and lifestyle characteristics. The Special Studies Institutional Review Board of the US National Cancer Institute approved the study (21). Participants gave informed consent by completing and returning the baseline questionnaire.

We sequentially excluded questionnaires filled out by proxy (n = 15 760); male participants (n = 325 171); any cancer diagnosis except nonmelanoma skin cancer prior to study entry (n = 23 925); cancers identified through death reports only (n = 1799); bilateral oophorectomy, oophorectomy, or unknown oophorectomy status (n = 47 790); implausibly high or low energy intake (more than 2 interquartile ranges above the 75th percentile or below the 25th percentile on the logarithmic scale; n = 1294) (22); and person-time equal to zero (n = 16). Our analytic cohort included 150 643 participants.

Prediagnosis diet quality

Participants completed a validated 124-item food frequency questionnaire at baseline regarding dietary intake over the past year (21,22). Nutrient and energy estimates were calculated based on the US Department of Agriculture Survey Nutrient Database associated with the Continuing Survey for Food Intake by Individuals 1994-1996 and the Nutrition Data System for Research database (University of Minnesota Nutrition Coordinating Center, 2004).

We then calculated 3 diet quality scores. The HEI-2015 assesses adherence to the Dietary Guidelines for Americans 2015-2020 according to 13 components (total fruits, whole fruits, total vegetables, greens and beans, total protein foods, seafood and plant proteins, whole grains, dairy, fatty acids, refined grains, sodium, added sugar, and saturated fats) and ranges from 0 to 100 (23). The HEI-2015 and the latest HEI-2020 are equivalent (24). The alternate Mediterranean diet (aMED) score measures alignment with the alternative Mediterranean diet (25) based on 9 components (vegetables excluding potatoes, fruits, nuts, whole grains, legumes, fish, monounsaturated-to-saturated fat ratio, red and processed meats, and alcohol) and ranges from 0 to 9 (25). The Dietary Approaches to Stop Hypertension (DASH) score assesses adherence to the DASH diet, which was designed for hypertension prevention (26), via 8 components (fruits, vegetables, nuts and legumes, whole grains, low-fat dairy products, sodium, red and processed meats, and sweetened beverages) and ranges from 8 to 40 (26). Higher scores represent higher diet quality and better adherence for all 3 scores. We created quintiles for each diet quality score based on baseline scores.

Outcome ascertainment

For the risk analysis, the primary outcome was first primary epithelial ovarian cancer ascertained through December 31, 2017, via linkage to 11 state cancer registries (8 original recruitment sites and Arizona, Nevada, and Texas). Epithelial ovarian cancer diagnoses were identified using the International Classification of Diseases for Oncology, Third Edition (C56.9) and morphology codes. Cases were classified into serous, endometrioid, mucinous, clear cell, and other epithelial ovarian cancers (Supplementary Table 1, available online). For the survival analysis among ovarian cancer patients, the primary outcome was all-cause mortality obtained through linkage with the National Death Index Plus through December 31, 2017.

Statistical analysis

We compared baseline characteristics across quintiles of HEI-2015, aMED, and DASH using analysis of variance or χ² tests in the full analytic sample and among the ovarian cancer patients. We only presented the highest and lowest quintiles in tables because of limited space.

Because ovarian cancer occurrence can be precluded by death, but not vice versa, we treated death as a semicompeting event (27). A semi-Markov multistate model (28) with multivariable Cox proportional hazard regression was used to model the transitions from baseline to epithelial ovarian cancer diagnosis (T1), from baseline to death (T2), and from epithelial ovarian cancer diagnosis to death (T3) (see Figure 1). We used the clock-forward method and further adjusted for age at cancer diagnosis in T3 to satisfy the Markov property. Using this model is important because ignoring the semicompeting risk of death can lead to biased results (27). We report hazard ratios and 95% confidence intervals for quintiles of the diet quality scores in relation to each transition and aggregate P values for diet quality scores from likelihood ratio tests. The linearity assumption was violated, thus we only modeled dietary scores as categorical variables.

Figure 1.

Semi-Markov multistate model system. The rectangles indicate different states. All participants of the analytic cohort enter the baseline enrollment state at baseline. The arrows between the rectangles indicate the transitions between states, such as T1 (development of first primary epithelial ovarian cancer), T2 (death without development of epithelial ovarian cancer), and T3 (death among people who were diagnosed with epithelial ovarian cancer).

Based on previous research on diet quality and ovarian cancer risk and survival, potential confounders were selected a priori for the ovarian cancer risk and the survival models. Baseline age, race and ethnicity, residency, education level, marriage status, family history of any cancer except for nonmelanoma skin cancers, hormone replacement therapy (HRT) use, oral contraceptives, comorbidities, leisure-time physical activity, smoking, body mass index (BMI), and total energy intake (kcal per day) were included in the risk and survival models. In the risk model, we additionally adjusted for number of livebirths, age at menarche, and menopausal status. In the survival model, we included cancer stage, cancer grade, and cancer histologic type.

The proportionality assumption for the Cox models was satisfied for HEI-2015 and DASH but was violated for aMED. However, results stratified by time were similar to nonstratified results for aMED; therefore, we only present nonstratified results.

We assessed potential effect modification for ovarian cancer risk by baseline age, HRT use, BMI, diabetes, and leisure-time physical activity with cross-product interaction terms in T1 and compared full and reduced models using Likelihood ratio tests. We also assessed potential effect modification for the survival analyses by age at diagnosis, HRT use, BMI, diabetes, leisure-time physical activity, cancer stage, histologic type, and cancer grade with cross-product interaction terms in T3 and Likelihood-ratio tests.

We conducted sensitivity analyses for ovarian cancer risk and survival excluding people who were diagnosed with cancer within the first 5 years of follow-up. We also conducted a sensitivity analysis for survival for cancer-specific deaths only. We further explored the relationships between HEI components and ovarian cancer survival to align with a prior publication (11).

We used SAS version 9.4 (SAS, Cary, NC, USA) for descriptive analyses. The semi-Markov models were implemented with survival and mstate packages in R Version 4.2.3 (28). Statistical tests were 2-sided with a .05 statistical significance level.

Results

The 150 643 participants contributed 2 444 594 person-years. The median follow-up time was 20.5 years, and 1107 people were diagnosed with first primary epithelial ovarian cancer. The median time from baseline to ovarian cancer diagnosis was 8.8 years.

Individuals in the highest quintile of the diet quality scores were generally older and better educated and more likely to be physically active and nonsmokers, as well as had a family history of cancer, fewer children, earlier menarche, used HRT, never taken oral contraceptive, fewer comorbidities, and had a BMI within the normal range (Table 1).

Table 1.

Baseline characteristics by diet quality score quintiles (n = 150 563)

Characteristics	Total No. (%) (n = 150 643)	HEI-2015 No. (%)		aMED No. (%)		DASH No. (%)
		Quintile 1 (n = 30 128)	Quintile 5 (n = 30 128)	Quintile 1 (n = 26 859)	Quintile 5 (n = 36 645)	Quintile 1 (n = 31 909)	Quintile 5 (n = 29 923)
Median (IQR) of diet quality score		56.0 (7.1)	80.0 (4.1)	2.0 (1.0)	6.0 (1.0)	19.0 (3.0)	29.0 (3.0)
Age, mean (SD), y	61.2 (5.4)	60.5 (5.5)	62.0 (5.3)	61.0 (5.5)	61.3 (5.4)	60.5 (5.5)	61.7 (5.4)
Race and ethnicity
American Indian or Alaskan Native	488 (0.3)	109 (0.4)	92 (0.3)	88 (0.3)	123 (0.3)	131 (0.4)	88 (0.3)
Asian	1762 (1.2)	290 (1.0)	313 (1.0)	292 (1.1)	484 (1.3)	417 (1.3)	349 (1.2)
Hispanic	2855 (1.9)	459 (1.5)	582 (1.9)	422 (1.6)	683 (1.9)	571 (1.8)	648 (2.2)
Non-Hispanic Black	7983 (5.3)	1453 (4.8)	1785 (5.9)	1062 (4.0)	2439 (6.7)	2085 (6.5)	1390 (4.7)
Non-Hispanic White	135 337 (89.8)	27 385 (90.9)	26 929 (89.4)	24 597 (91.6)	32 362 (88.3)	28 170 (88.3)	26 975 (90.2)
Pacific Islander	209 (0.1)	37 (0.1)	44 (0.2)	37 (0.1)	66 (0.2)	50 (0.2)	42 (0.1)
Unknown	2009 (1.3)	395 (1.3)	382 (1.3)	361 (1.3)	488 (1.3)	485 (1.5)	431 (1.4)
Residency at enrollment
California	48 547 (32.2)	7762 (25.8)	10 962 (36.4)	7803 (29.1)	12 795 (34.9)	8708 (27.3)	11 111 (37.1)
Florida	30 618 (20.3)	6088 (20.2)	6035 (20.0)	5480 (20.4)	7208 (19.7)	6889 (21.6)	5543 (18.5)
Atlanta, Georgia	4167 (2.8)	760 (2.5)	833 (2.8)	743 (2.8)	1031 (2.8)	961 (3.0)	759 (2.5)
Louisiana	5116 (3.4)	1141 (3.8)	948 (3.2)	982 (3.7)	1146 (3.1)	1345 (4.2)	758 (2.5)
Detroit, Michigan	7810 (5.2)	1634 (5.4)	1566 (5.2)	1494 (5.6)	1850 (5.1)	1775 (5.6)	1413 (4.7)
North Carolina	12 233 (8.1)	2401 (8.0)	2489 (8.3)	1981 (7.4)	3103 (8.5)	3025 (9.5)	2179 (7.3)
New Jersey	19 598 (13.0)	4731 (15.7)	3478 (11.5)	3770 (14.0)	4534 (12.4)	4051 (12.7)	4008 (13.4)
Pennsylvania	22 554 (15.0)	5611 (18.6)	3817 (12.7)	4606 (17.2)	4978 (13.6)	5155 (16.2)	4152 (13.9)
Education level
<12 years	8614 (5.7)	2527 (8.4)	1196 (4.0)	1961 (7.3)	1630 (4.5)	2542 (7.9)	1226 (4.1)
12 years	37 532 (24.9)	9510 (31.6)	6017 (20.0)	8.157 (30.4)	7392 (20.2)	9854 (30.9)	5717 (19.1)
College or post-high school training	52 989 (35.2)	10 406 (34.5)	10 463 (34.7)	9430 (35.1)	12 628 (34.5)	11 286 (35.4)	10 192 (34.1)
Advanced degree	46 035 (31.2)	6758 (22.4)	116 018 (38.6)	6445 (24.0)	13 895 (37.9)	7219 (22.6)	11 862 (39.6)
Unknown	4473 (3.0)	927 (3.1)	834 (2.8)	866 (3.2)	1110 (3.0)	1008 (3.2)	926 (3.1)
Marital status
Married or living as married	66 581 (44.2)	12 397 (41.2)	13 638 (45.3)	11 052 (41.2)	17 138 (46.8)	14 552 (45.6)	12 365 (41.3)
Widowed	33 447 (22.2)	7056 (23.4)	6667 (22.1)	6576 (24.5)	7483 (20.4)	6984 (21.9)	6665 (22.3)
Divorced	36 508 (24.2)	7587 (25.2)	7120 (23.6)	6622 (24.7)	8743 (23.9)	7320 (22.9)	8004 (26.8)
Separated	2382 (1.6)	569 (1.9)	426 (1.4)	415 (1.6)	570 (1.6)	540 (1.7)	478 (1.6)
Never married	10 478 (7.0)	2274 (7.5)	2056 (6.8)	1941 (7.2)	2427 (6.6)	2218 (7.0)	2152 (7.2)
Unknown	1247 (0.8)	245 (0.8)	221 (0.7)	253 (0.9)	284 (0.8)	295 (0.9)	259 (0.9)
No. of live-born children
0	22 932 (15.2)	4739 (15.7)	4661 (15.5)	4043 (15.1)	5611 (15.3)	4711 (14.8)	4808 (16.1)
1-2	54 522 (36.2)	10 638 (35.3)	11 119 (36.9)	9512 (35.4)	13 439 (36.7)	11 391 (35.7)	11 096 (37.1)
>2	72 603 (48.2)	14 622 (48.5)	14 249 (47.3)	13 183 (49.1)	17 467 (47.7)	15 657 (49.1)	13 913 (46.5)
Unknown	586 (0.4)	129 (0.4)	99 (0.3)	121 (0.5)	128 (0.4)	150 (0.5)	106 (0.4)
Age at menarche, y
10 or younger	9924 (6.6)	1993 (6.6)	1945 (6.5)	1688 (6.3)	2442 (6.7)	2059 (6.5)	2064 (6.9)
11-12	62 690 (41.6)	12 108 (40.2)	12 896 (42.8)	11 122 (41.4)	15 170 (41.4)	12 952 (40.6)	12 470 (41.7)
13-14	63 222 (42.0)	12 702 (42.2)	12 674 (42.1)	11 240 (41.9)	15 528 (42.4)	13 357 (41.9)	12 558 (42.0)
15 or older	14 335 (9.5)	3213 (10.7)	2531 (8.4)	2701 (10.1)	3410 (9.3)	3408 (10.7)	2733 (9.1)
Unknown	472 (0.3)	112 (0.4)	82 (0.3)	108 (0.4)	95 (0.3)	133 (0.4)	98 (0.3)
Postmenopausal	142 145 (94.4)	28 340 (94.1)	28 638 (95.1)	25 345 (94.4)a	34 570 (94.3)a	29 996 (94.0)	28 360 (94.8)
Family history of cancer
No	60 106 (39.9)	11 370 (37.7)	12 631 (41.9)	10 234 (38.1)	15 240 (41.6)	11 908 (37.3)	12 551 (41.9)
Yes	30 486 (20.2)	5745 (19.1)	6569 (21.8)	5276 (19.6)	7682 (21.0)	6054 (19.0)	6481 (21.7)
Unknown	60 051 (39.9)	13 013 (43.2)	10 928 (36.3)	11 349 (42.3)	13 723 (37.5)	13 947 (43.7)	10 891 (36.4)
Ever taken hormone replacement therapy	70 061 (46.5)	11 870 (39.4)	15 487 (51.4)	11 410 (42.5)	17 988 (49.1)	13 281 (41.6)	14 770 (49.4)
Oral contraceptive
Never	88 564 (58.8)	17 562 (58.3)	18 345 (60.9)	15 763 (58.7)a	21 496 (58.7)a	18 152 (56.9)	18 298 (61.2)
1-4 years	27 023 (17.9)	5423 (18.0)	5183 (17.2)	4833 (18.0)a	6685 (18.2)a	5865 (18.4)	5175 (17.3)
5-9 years	19 170 (12.7)	3862 (12.8)	3622 (12.0)	3414 (12.7)a	4630 (12.6)a	4279 (13.4)	3502 (11.7)
≥10	14 830 (9.8)	3063 (10.2)	2779 (9.2)	2629 (9.8)a	3600 (9.8)a	3349 (10.5)	2749 (9.2)
Unknown	1056 (0.7)	218 (0.7)	199 (0.7)	220 (0.8)a	234 (0.6)a	264 (0.8)	199 (0.7)
No. of comorbidities
0	93 182 (61.9)	18 033 (59.9)	19 015 (63.1)	16 228 (60.4)	23 361 (63.8)	19 583 (61.4)	19 010 (63.5)
1-2	53 165 (35.3)	11 076 (36.8)	10 347 (34.3)	9746 (36.3)	12 388 (33.8)	11 330 (35.5)	10 131 (33.9)
≥3	4296 (2.9)	1019 (3.4)	766 (2.5)	885 (3.3)	896 (2.5)	996 (3.1)	782 (2.6)
Smoking
Never	66 708 (44.3)	11 696 (38.8)	14 356 (47.6)	11 099 (41.3)	17 051 (46.5)	12 589 (39.5)	14 334 (47.9)
Former	58 149 (38.6)	9884 (32.8)	12 793 (42.5)	9300 (34.6)	15 210 (41.5)	10 682 (33.5)	12 553 (42.0)
Current	21 441 (14.2)	7652 (25.4)	2161 (7.2)	5680 (21.2)	3323 (9.1)	7738 (24.3)	2129 (7.1)
Unknown	4345 (2.9)	896 (3.0)	818 (2.7)	780 (2.9)	1061 (2.9)	900 (2.8)	907 (3.0)
Leisure time physical activity, mean (SD), h per wk	4.3 (5.7)	3.1 (4.8)	5.4 (6.1)	3.3 (4.8)	5.3 (6.3)	3.1 (5.7)	5.7 (6.5)
Body mass index, kg/m2
<18.5	2314 (1.5)	599 (2.0)	405 (1.3)	446 (1.7)	544 (1.5)	288 (1.8)	472 (1.6)
18.5 to <25	63 955 (42.5)	11 428 (38.0)	14 674 (48.7)	10 523 (39.2)	16 884 (46.1)	12 300 (38.6)	14 618 (48.9)
25 to <30	46 351 (30.8)	8875 (29.5)	9103 (30.2)	8545 (31.8)	10 906 (29.8)	9867 (30.9)	8586 (28.7)
≥30	33 410 (22.2)	8141 (27.0)	5104 (16.9)	6422 (23.9)	7262 (19.8)	8027 (25.2)	5373 (18.0)
Unknown	4613 (3.1)	1085 (3.6)	842 (2.8)	923 (3.4)	1049 (2.9)	1127 (3.5)	874 (2.9)
Total energy intake, mean (SD), kcal/day	1570.7 (653.4)	1686.7 (749.7)	1486.4 (561.6)	1193.3 (502.8)	1932.3 (683.1)	1412.8 (613.5)	1787.5 (669.0)

a

P value (for analysis of variance or χ² test) are not statistically significant at .05 level; all other P values less than .03. aMED = Alternative Mediterranean Diet score; DASH = Dietary Approaches to Stop Hypertension score; HEI-2015 = Healthy Eating Index-2015; IQR = interquartile range.

The mean (SD) HEI-2015, aMED, and DASH scores were 68.76 (9.36), 4.20 (1.74), and 23.90 (4.11), respectively, with moderate correlations between the 3 (Pearson correlation coefficients: HEI-2015 and aMED = 0.50; HEI-2015 and DASH = 0.62; aMED and DASH = 0.61). Individual components of the diet quality scores are presented in Supplementary Table 2 (available online).

Among those diagnosed with epithelial ovarian cancer, 893 died. The median survival after ovarian cancer diagnosis was 2.5 years (Table 2). People who reported higher diet quality scores were more likely to be diagnosed later in life, but stage, grade, and histology were similar across the diet quality quintiles.

Table 2.

Baseline and clinical characteristics of people diagnosed with ovarian cancer by diet quality score quintiles (n = 1107)

Characteristics	Total No. (%) (n = 1107)	HEI-2015 No. (%)		aMED No. (%)		DASH No. (%)
		Quintile 1 (n = 193)	Quintile 5 (n = 228)	Quintile 1 (n = 183)	Quintile 5 (n = 278)	Quintile 1 (n = 221)	Quintile 5 (n = 204)
Median (IQR) of diet quality scorea		57.1 (5.4)	79.9 (3.1)	2.0 (0.6)	6.0 (0.7)	19.0 (1.9)	30.0 (1.7)
Age at diagnosis, mean (SD), y	71.0 (7.4)	70.2 (7.4)a	71.9 (7.2)a	70.7 (7.5)	71.5 (7.5)	70.3 (7.5)a	71.3 (7.1)a
Race and ethnicity
American Indian or Alaskan Native	3 (0.3)	2 (1.0)	0 (0.0)	1 (0.6)	0 (0.0)	2 (0.9)	0 (0.0)
Asian	5 (0.5)	1 (0.5)	2 (0.9)	2 (1.1)	2 (0.7)	2 (0.9)	2 (1.0)
Hispanic	20 (1.8)	3 (1.6)	4 (1.8)	2 (1.1)	3 (1.1)	5 (2.3)	4 (2.0)
Non-Hispanic Black	40 (3.6)	5 (2.6)	8 (3.5)	4 (2.2)	10 (3.6)	9 (4.1)	7 (3.4)
Non-Hispanic White	1016 (91.8)	180 (93.3)	212 (93.0)	170 (92.9)	259 (93.2)	197 (89.1)	187 (91.7)
Pacific Islander	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Unknown	23 (2.1)	2 (1.0)	2 (0.9)	4 (2.2)	4 (1.4)	6 (2.7)	4 (2.0)
Residency at enrollment
California	404 (36.5)	56 (29.0)	94 (41.2)	59 (32.2)	108 (38.9)	77 (34.8)	83 (40.7)
Florida	215 (19.4)	37 (19.2)	42 (18.4)	31 (16.9)	53 (19.1)	44 (19.9)	34 (16.7)
Atlanta, Georgia	35 (3.2)	3 (1.6)	9 (4.0)	4 (2.2)	10 (3.6)	4 (1.8)	6 (2.9)
Louisiana	45 (4.1)	10 (5.2)	5 (2.2)	11 (6.0)	6 (2.2)	15 (6.8)	7 (3.4)
Detroit, Michigan	2 (0.2)	0 (0.0)	0 (0.0)	0 (0.0)	1 (0.4)	0 (0.0)	0 (0.0)
North Carolina	94 (8.5)	15 (7.8)	23 (10.1)	15 (8.2)	30 (10.8)	18 (8.1)	21 (10.3)
New Jersey	135 (12.2)	34 (17.6)	23 (10.1)	24 (13.1)	34 (12.2)	27 (12.2)	21 (10.3)
Pennsylvania	177 (16.0)	38 (19.7)	32 (14.0)	39 (21.3)	36 (13.0)	36 (16.3)	32 (15.7)
Education levela
<12 years	53 (4.8)	10 (5.20	10 (4.4)	10 (5.5)	6 (2.2)	17 (7.7)	7 (3.4)
12 years	264 (23.9)	62 (32.1)	39 (17.1)	59 (32.2)	42 (15.1)	63 (28.5)	31 (15.2)
College or post-high school training	391 (35.3)	70 (36.3)	83 (36.4)	58 (31.7)	93 (33.5)	82 (37.1)	69 (33.8)
Advanced degree	366 (33.0)	48 (24.8)	89 (39.0)	50 (27.3)	129 (46.4)	53 (24.0)	90 (44.1)
Unknown	33 (3.0)	3 (1.6)	7 (3.1)	6 (3.3)	8 (2.9)	6 (2.7)	7 (3.4)
Marital status
Married or living as married	501 (45.3)	82 (42.5)	99 (43.4)	79 (43.2)	115 (41.4)	114 (51.6)a	79 (38.7)a
Widowed	220 (19.9)	36 (18.7)	50 (21.9)	34 (18.6)	61 (21.9)	36 (16.3)a	50 (24.5)a
Divorced	268 (24.2)	50 (25.9)	49 (21.5)	47 (25.7)	75 (27.0)	43 (19.5)a	49 (24.0)a
Separated	13 (1.2)	3 (1.6)	4 (1.8)	2 (1.1)	3 (1.1)	5 (2.3)a	4 (2.0)a
Never married	101 (9.1)	21 (10.9)	25 (11.0)	21 (11.5)	23 (8.3)	23 (10.4)a	22 (10.8)a
Unknown	4 (0.4)	1 (0.5)	1 (0.4)	0 (0.0)	1 (0.4)	0 (0.0)a	0 (0.0)a
No. of live-born children
0	211 (19.1)	48 (24.9)	48 (21.1)	40 (21.9)	51 (18.4)	48 (21.7)	46 (22.6)
1-2	427 (38.6)	69 (35.8)	83 (36.4)	74 (40.4)	117 (42.1)	81 (36.7)	70 (34.3)
>2	466 (42.2)	74 (38.3)	97 (42.5)	69 (37.7)	110 (39.6)	92 (41.6)	88 (43.1)
Unknown	3 (0.3)	2 (1.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)	0 (0.0)
Age at menarche, y
10 or younger	55 (5.0)	11 (5.7)	9 (4.0)	11 (6.0)	14 (5.0)	10 (4.5)	11 (5.4)
11-12	475 (42.9)	79 (40.9)	102 (44.7)	67 (36.6)	115 (41.4)	90 (40.7)	91 (44.6)
13-14	474 (42.8)	76 (39.4)	94 (41.2)	85 (46.5)	119 (42.8)	100 (45.3)	84 (41.2)
15 or older	98 (8.9)	26 (13.5)	22 (9.7)	18 (9.8)	29 (10.4)	18 (8.1)	16 (7.8)
Unknown	5 (0.5)	1 (0.5)	1 (0.4)	2 (1.1)	1 (0.4)	3 (1.4)	2 (1.0)
Postmenopausal	1061 (95.8)	184 (95.3)	220 (96.5)	175 (95.6)	268 (96.4)	215 (97.3)	198 (97.1)
Family history of cancer
No	418 (37.8)	66 (34.2)	32 (40.8)	62 (33.9)	112 (40.3)	70 (31.7)a	87 (42.7)a
Yes	235 (21.3)	33 (17.1)	55 (24.1)	40 (21.9)	63 (22.7)	39 (17.7)a	47 (23.0)a
Unknown	454 (41.0)	94 (48.7)	80 (35.1)	81 (44.3)	103 (37.1)	112 (50.7)a	70 (34.3)a
Ever taken hormone replacement therapya	592 (53.5)	79 (40.9)	140 (61.4)	70 (43.2)	160 (57.6)	99 (44.8)	120 (58.8)
Oral contraceptive, y
Never	731 (66.0)	117 (60.6)	149 (65.4)	133 (72.7)	186 (66.9)	145 (65.6)	137 (67.2)
1-4	186 (16.8)	35 (18.1)	36 (15.8)	28 (15.3)	48 (17.3)	39 (17.7)	34 (16.7)
5-9	115 (10.4)	26 (13.5)	23 (10.1)	11 (6.0)	26 (9.4)	22 (10.0)	18 (8.8)
≥10	68 (6.1)	13 (6.7)	17 (7.5)	10 (5.5)	16 (5.8)	13 (5.9)	13 (6.4)
Unknown	7 (0.6)	2 (1.0)	3 (1.3)	1 (0.6)	2 (0.7)	2 (0.9)	2 (1.0)
No. of comorbidities
0	685 (61.9)	109 (56.5)	150 (65.8)	104 (56.8)	185 (66.6)	121 (54.8)	128 (62.8)
1-2	394 (35.6)	78 (40.4)	74 (32.5)	75 (41.0)	89 (32.0)	91 (41.2)	71 (34.8)
≥3	28 (2.5)	6 (3.1)	4 (1.8)	4 (2.2)	4 (1.4)	9 (4.1)	5 (2.5)
Smoking
Never	560 (50.6)	82 (42.5)a	113 (49.6)a	102 (55.7)	144 (51.8)	112 (50.7)a	115 (56.4)a
Former	409 (37.0)	64 (33.2)a	103 (45.2)a	50 (27.3)	106 (38.1)	72 (32.6)a	73 (35.8)a
Current	113 (10.2)	41 (21.2)a	10 (4.4)a	26 (14.2)	21 (7.6)	31 (14.0)a	10 (4.9)a
Unknown	25 (2.3)	6 (3.1)a	2 (0.9)a	5 (2.7)	7 (2.5)	6 (2.7)a	6 (2.9)a
Leisure time physical activity, mean (SD), h/wk	4.7 (6.6)	4.7 (8.5)a	5.5 (6.5)a	3.7 (6.0)	5.5 (7.7)	4.0 (6.7)a	7.0 (9.4)a
Body mass index, kg/m2a
<18.5	15 (1.4)	3 (1.6)	1 (0.5)	3 (1.6)	1 (0.4)	6 (2.7)	0 (0.0)
18.5 to <25	477 (43.1)	69 (35.8)	118 (51.8)	70 (38.3)	139 (50.0)	77 (34.8)	102 (50.0)
25 to <30	330 (29.8)	53 (27.5)	65 (28.5)	52 (28.4)	83 (29.9)	65 (29.4)	57 (27.9)
≥30	245 (22.1)	60 (31.1)	36 (15.4)	48 (26.3)	48 (17.3)	65 (29.4)	37 (18.1)
Unknown	40 (3.6)	8 (4.2)	9 (4.0)	10 (5.5)	7 (2.5)	8 (3.6)	8 (3.9)
Cancer stage
Localized	56 (5.1)	12 (6.2)	12 (5.3)	11 (6.0)	13 (4.7)	12 (5.4)	11 (5.4)
Regional	119 (10.8)	25 (13.0)	21 (9.2)	19 (10.4)	23 (8.3)	29 (13.1)	19 (9.3)
Distant	624 (56.4)	112 (58.0)	126 (55.3)	99 (54.1)	169 (60.8)	118 (53.4)	108 (52.9)
Unknown	308 (27.8)	44 (22.8)	69 (30.3)	54 (29.5)	73 (26.3)	62 (28.1)	66 (32.4)
Cancer grade
I	44 (4.0)	10 (5.2)	10 (4.4)	6 (3.3)	10 (3.6)	8 (3.6)	7 (3.4)
II	148 (13.4)	26 (13.5)	33 (14.5)	19 (10.4)	42 (15.1)	27 (12.2)	26 (12.8)
III	449 (40.6)	89 (46.1)	91 (39.9)	88 (48.1)	111 (39.9)	91 (41.2)	84 (41.2)
IV	128 (11.6)	16 (8.3)	28 (12.3)	14 (7.7)	39 (14.0)	19 (8.6)	28 (13.7)
Unknown	338 (30.5)	52 (26.9)	66 (29.0)	56 (30.6)	76 (27.3)	76 (34.4)	59 (28.9)
Histologic type
Serous	633 (57.2)	108 (56.0)	133 (58.3)	107 (58.5)	162 (58.3)	113 (51.1)	122 (59.8)
Endometrioid	92 (8.3)	11 (5.7)	21 (9.2)	14 (7.7)	29 (10.4)	19 (8.6)	17 (8.3)
Mucinous	50 (4.5)	13 (6.7)	9 (4.0)	7 (3.8)	9 (3.2)	11 (5.0)	9 (4.4)
Clear cell	24 (2.2)	6 (3.1)	2 (0.9)	7 (3.8)	3 (1.1)	9 (4.1)	2 (1.0)
Other epithelial	308 (27.8)	55 (28.5)	63 (27.6)	48 (26.2)	75 (27.0)	69 (31.2)	54 (26.5)

a

P value (for analysis of variance, χ² test, or Fisher exact test) less than .05; all other P values greater than .05. aMED = Alternative Mediterranean Diet score; DASH = Dietary Approaches to Stop Hypertension score; HEI-2015 = Healthy Eating Index-2015; IQR = interquartile range.

There was no evidence of an association between diet quality (HEI-2015, aMED, or DASH) and ovarian cancer risk (Table 3). We also did not observe any evidence of effect modification for risk (P_interactions > .05; data not shown).

Table 3.

Associations between diet quality and epithelial ovarian cancer risk (n = 150 643)

Dietary scores	Quintile 1	Quintile 2	Quintile 3	Quintile 4	Quintile 5	Paggregatec
HEI-2015
Median (IQR) of diet score	55.99 (7.13)	64.52 (3.05)	69.76 (2.35)	74.37 (2.35)	79.98 (4.11)
Cases/person-years	193/476 074.9	249/485 682.8	214/489 486.5	223/494 698.5	228/498 651.6
Age-adjusted HR (95% CI)a	Referent	1.25 (1.04 to 1.51)	1.07 (0.88 to 1.30)	1.09 (0.90 to 1.32)	1.09 (0.90 to 1.32)	.29
Multivariable-adjusted HR (95% CI)b	Referent	1.22 (1.01 to 1.48)	1.03 (0.85 to 1.26)	1.04 (0.85 to 1.27)	1.03 (0.84 to 1.26)	.20
aMED
Median (IQR) of diet score	2.00 (1.00)	3.00 (0.00)	4.00 (0.00)	5.00 (0.00)	6.00 (1.00)
Cases/person-years	183/426 196.9	184/428 742.7	236/500 963.9	226/481 042.8	278/607 647.6
Age-adjusted HR (95% CI)a	Referent	1.00 (0.82 to 1.23)	1.10 (0.91 to 1.34)	1.11 (0.91 to 1.36)	1.09 (0.89 to 1.33)	.73
Multivariable-adjusted HR (95% CI)b	Referent	0.98 (0.80 to 1.21)	1.07 (0.88 to 1.30)	1.07 (0.87 to 1.31)	1.03 (0.84 to 1.27)	.88
DASH
Median (IQR) of diet score	19.00 (3.00)	22.00 (1.00)	24.00 (2.00)	26.00 (1.00)	29.00 (3.00)
Cases/person-years	221/507 760.7	180/379 529.5	325/664 955.9	177/397 235.4	204/495 112.4
Age-adjusted HR (95% CI)a	Referent	1.07 (0.88 to 1.31)	1.10 (0.93 to 1.31)	1.00 (0.82 to 1.22)	0.92 (0.76 to 1.12)	.32
Multivariable-adjusted HR (95% CI)b	Referent	1.04 (0.86 to 1.27)	1.05 (0.88 to 1.25)	0.92 (0.75 to 1.13)	0.83 (0.68 to 1.02)	.10

a

Hazard ratio (95% CI) estimated using Cox regression adjusted for baseline age and total energy intake (kcal/day). aMED = Alternative Mediterranean Diet score; CI = confidence interval; DASH = Dietary Approaches to Stop Hypertension score; HEI-2015 = Healthy Eating Index-2015; HR = hazard ratio.

b

Hazard ratio (95% CI) estimated using Cox regression adjusted for baseline age, race and ethnicity (non-Hispanic White; non-Hispanic Black; Hispanic; Asian, Pacific Islander, American Indian, or Alaskan Native; unknown), residency (California; Florida; Louisiana; New Jersey; North Carolina; Pennsylvania; Atlanta, Georgia; Detroit, Michigan), education level (<12 years, 12 years or completed high school, post-high school training or some college, college or advanced degree, unknown), marriage status (married or living as married, widowed, divorced, separated, never married, unknown), number of liveborn (0, 1-2, >2, unknown), age at menarche (10 years of younger, 11-12 years, 13-14 years, 15 years or older), postmenopausal (yes, no), family history of any cancer (yes, no, unknown), hormone replacement therapy usage (ever, never), oral contraceptives (never, 1-4 years, 5-9 years, 10 or more years, unknown), comorbidities (0, 1-2, ≥3), leisure-time physical activity (tertiles), smoking (never, former, current), body mass index (kg/m²; <18.5, 18.5 to <25, 25 to <30, ≥30, unknown), and total energy intake (kcal/day).

c

Aggregate P value for diet quality score from likelihood ratio test can be interpreted as the significance of each diet quality score as a whole.

Among people diagnosed with ovarian cancer, the highest quintile of prediagnosis HEI-2015 score was associated with better survival compared with the lowest quintile (Q5 vs Q1: HR = 0.75, 95% CI = 0.60 to 0.93) (Table 4). Similarly, the highest quintile of aMED was associated with better survival compared with the lowest quintile (HR = 0.68, 95% CI = 0.53 to 0.87). DASH was not associated with all-cause mortality.

Table 4.

Association between diet quality and all-cause mortality among those diagnosed with epithelial ovarian cancer (n = 1107)

Dietary scores	Quintile 1	Quintile 2	Quintile 3	Quintile 4	Quintile 5	Paggregatec
HEI-2015
Cases/person-years	164/795.9	199/1222.3	175/1018.9	183/1054.8	172/1130.9
Age-adjusted HR (95% CI)a	Referent	0.76 (0.62 to 0.93)	0.77 (0.62 to 0.95)	0.74 (0.60 to 0.92)	0.66 (0.53 to 0.81)	<.01
Multivariable-adjusted HR (95% CI)b	Referent	0.75 (0.60 to 0.93)	0.71 (0.57 to 0.90)	0.73 (0.58 to 0.92)	0.75 (0.59 to 0.95)	.04
aMED
Cases/person-years	150/844.1	145/773.1	193/1097.6	189/1086.9	216/1420.7
Age-adjusted HR (95% CI)a	Referent	0.99 (0.79 to 1.25)	0.95 (0.76 to 1.17)	0.89 (0.71 to 1.11)	0.71 (0.56 to 0.89)	.01
Multivariable-adjusted HR (95% CI)b	Referent	0.93 (0.73 to 1.18)	0.98 (0.78 to 1.23)	0.91 (0.71 to 1.16)	0.68 (0.53 to 0.87)	<.01
DASH
Cases/person-years	170/1034.4	152/827.4	266/1589.4	143/819.3	162/951.9
Age-adjusted HR (95% CI)a	Referent	1.11 (0.89 to 1.38)	0.86 (0.71 to 1.05)	0.85 (0.68 to 1.07)	0.88 (0.71 to 1.10)	.09
Multivariable-adjusted HR (95% CI)b	Referent	0.99 (0.79 to 1.25)	0.88 (0.71 to 1.07)	0.85 (0.66 to 1.08)	1.01 (0.80 to 1.29)	.35

a

Hazard ratio (95% CI) estimated using Cox regression adjusted for age at diagnosis and total energy intake (kcal/day). aMED = Alternative Mediterranean Diet score; CI = confidence interval; DASH = Dietary Approaches to Stop Hypertension score; HEI-2015 = Healthy Eating Index-2015; HR = hazard ratio.

b

Hazard ratio (95% CI) estimated using Cox regression adjusted for age at diagnosis, race and ethnicity (non-Hispanic White; non-Hispanic Black; Hispanic; Asian, Pacific Islander, American Indian, or Alaskan Native; unknown), residency (California; Florida; Louisiana; New Jersey; North Carolina; Pennsylvania; Atlanta, Georgia; Detroit, Michigan), education level (<12 years, 12 years or completed high school, post-high school training or some college, college or advanced degree, unknown), marriage status (married or living as married, widowed, divorced, separated, never married, unknown), family history of any cancer (yes, no, unknown), hormone replacement therapy usage (ever, never), oral contraceptives (never, 1-4 years, 5-9 years, ≥10 years, unknown), comorbidities (0, 1-2, ≥3), leisure-time physical activity (tertiles), smoking (never, former, current), body mass index (kg/m²; <18.5, 18.5 to <25, 25 to <30, ≥30, unknown), cancer stage (localized, regional, distant, unknown), cancer grade (I, II, III, IV, unknown), cancer histologic type (serous, endometrioid, mucinous, clear cell, other epithelial), and total energy intake (kcal/day).

c

Aggregate P value for diet quality score from likelihood ratio test can be interpreted as the significance of each diet quality score as a whole.

Disease stage modified the associations between dietary scores and survival (P_interaction: HEI-2015 = 0.046; aMED = 0.01; DASH < .01) (Supplementary Table 3, available online). Higher diet quality scores were associated with better survival for distant disease but not for people with unknown stage. The associations were in the opposite direction for localized/regional disease. Histologic type also modified the relationships between HEI-2015 (P_interaction = .03) and all-cause mortality (Supplementary Table 3, available online) but not aMED (P_interaction = .07) or DASH (P_interaction = .23). The inverse associations were stronger among nonserous epithelial ovarian cancer patients.

When we excluded individuals diagnosed with cancer within the first 5 years of follow-up, risk and survival findings were similar to our primary results (Supplementary Table 4, available online). When restricted to only cancer-specific deaths (n = 716), aMED was associated with better survival (Q5 vs Q1: HR = 0.74, 95% CI = 0.56 to 0.98), while the association between HEI-2015 and survival was no longer statistically significant, though the effect size was similar to our primary analysis (Q5 vs Q1: HR = 0.80, 95% CI = 0.62 to 1.05) (Supplementary Table 5, available online).

In our exploratory analysis of the HEI-2015 individual components, none were associated with ovarian cancer survival (Supplementary Table 6, available online).

Discussion

In this large prospective cohort, prediagnosis diet quality was inversely associated with ovarian cancer survival, with a 25% and 32% reduction in all-cause mortality accounting for competing risks for the highest quintile vs the lowest quintile for HEI-2015 and aMED, respectively. However, there was no evidence of an association between diet quality and ovarian cancer risk.

Three prospective cohort studies have assessed the association between diet quality and ovarian cancer survival (11-13). Similar to our effect size for HEI-2015, among 636 ovarian cancer patients in the Women’s Health Initiative (11), higher prediagnosis HEI-2005 was associated with lower mortality (highest vs lowest tertile: HR = 0.73, 95% CI = 0.55 to 0.97). The Ovarian Cancer Follow-Up Study (n = 796) found higher postdiagnosis Chinese HEI was associated with higher overall survival (highest vs lowest tertile: HR = 0.56, 95% CI = 0.36 to 0.88), but there was no association with the Dietary Balance Index or the Chinese Food Pagoda Score (13). Neither pre- (n = 650) nor postdiagnosis (n = 503) diet quality, assessed via HEI-2010, Alternate HEI-2010 (AHEI-2010), World Cancer Research Fund index, American Cancer Society Guidelines, or the Australian Dietary Guideline Index, was associated with ovarian cancer survival in the OPAL study (12), though the effect sizes were modest, and the statistical power could be limited by the relatively small sample size.

Although one might hypothesize diet quality would be associated with ovarian cancer risk and survival, similar to our study, 2 previous prospective cohorts failed to discover statistically significant relationships between diet quality and ovarian cancer risk (8,9). There was no statistically significant association between recommended foods score and ovarian cancer risk (highest vs lowest quartile: relative risk [RR] = 0.76, 95% CI = 0.47 to 1.22) among 42 254 participants (n = 142) in the prospective Breast Cancer Detection Demonstration Project cohort over 9.5 years (8), but there was likely limited power in this analysis given the limited number of patients even though the point estimate indicated a 24% lower risk. HEI-2005 (Q5 vs Q1: RR = 0.85, 95% CI = 0.65 to 1.12), AHEI-2010 (Q5 vs Q1: RR = 1.03, 95% CI = 0.80 to 1.34), and aMED (Q5 vs Q1: RR = 0.91, 95% CI = 0.71 to 1.18) were not associated with ovarian cancer risk among 82 948 participants (n = 696) in the Nurses’ Health Study during 24 years of follow-up (9). Case-control studies of diet quality are more mixed (6,7,10). One study of 205 ovarian cancer patients and 390 controls found no association between HEI-2005 and ovarian cancer risk (highest vs lowest tertile: odds ratio [OR] = 0.90, 95% CI = 0.55 to 1.47) (10). Another study with 415 ovarian cancer patients and 629 controls among African American women was null for HEI-2005 and HEI-2010 in relation to ovarian cancer risk, but higher AHEI-2010 score was related to lower ovarian cancer risk (highest vs lowest quartile: OR = 0.66, 95% CI = 0.45 to 0.98; P_trend = .05) (6). An Italian case-control study with 1031 ovarian cancer patients and 2411 controls found the Diabetes Risk Reduction Diet was inversely associated with ovarian cancer risk (highest vs lowest quartile: OR = 0.76, 95% CI = 0.60 to 0.95) (7). Although difficult to directly compare these results because diet quality measures have varied, all studies that used HEI and aMED scores were null for ovarian cancer risk (8,9), including ours. The inconsistencies across the case-control studies could also potentially be due to recall bias or dietary changes because of early symptoms (6,7).

Diet quality may impact ovarian cancer survival directly through biological pathways and indirectly through adherence to chemotherapy. Although mechanistic data for ovarian cancer survivors are limited, better diet quality has been associated with lower C-reactive protein levels in breast cancer survivors (29), and higher HEI scores were correlated with lower inflammation diets in a healthy population (30). Dietary quality could also affect cancer survival via metabolic pathways, such as glucose and lipid metabolism (20). In addition, among patients with solid tumors, better diet quality during cancer treatment has been related to better nutritional status, less muscle loss during treatment, and fewer chemotherapy side effects (31,32), which can lead to better survival directly (33,34) or through better adherence to chemotherapy (35). We only assessed prediagnosis diet, and we lack cancer treatment data, so we could not explore direct effects of diet during or after treatment on ovarian cancer outcomes or control for treatment-related characteristics. More studies, like Ovarian Cancer Follow-Up Study and OPAL, are warranted to assess postdiagnosis diet quality and health outcomes in ovarian cancer.

Although we did observe an interaction by disease stage for survival with unexpected adverse associations for higher diet quality and survival among those participants with localized/regional disease, only 15.9% of cases (n = 175) were localized/regional. Sparse cases in the diet quality score quintiles led to large variation in effect estimates. We also observed that the inverse association between HEI-2015 and mortality was stronger for nonserous epithelial ovarian cancer. In contrast, though not a diet quality score analysis, a previous cohort study with 811 participants noted a stronger association between higher intake of total vegetables, oily fish, and green tea and survival in high-grade serous epithelial cancer patients (36). Additional studies with more early stage ovarian cancer cases and studies stratified by histologic type are needed to further examine these potential interactions.

When restricted to cancer-specific mortality, the association between HEI-2015 and survival was similar to our primary analysis but no longer statistically significant. This finding could be because of the smaller number of cancer-specific deaths compared with all-cause deaths, or it might reflect how HEI-2010 had previously been more strongly inversely associated with all-cause mortality compared with cancer-specific mortality in women in the NIH-AARP Diet and Health cohort than was seen for aMED (37). We assessed cancer-specific deaths in this sensitivity analysis, but one could assume most of these deaths were ovarian cancer related.

We observed varying results across the 3 dietary scores, which is likely because of the different components included in each score. A previous study in the NIH-AARP Diet and Health Study found aMED was more strongly associated with cancer mortality in women than HEI-2010 and DASH (37). HEI-2010, which is based on the Dietary Guidelines for Americans 2010-2015, is similar to HEI-2015 except that HEI-2015 has saturated fat and added sugars in place of the empty calories component in HEI-2010 (23). Compared with aMED and DASH, HEI-2015 includes greater granularity for fatty acids, grains, and protein (24). Both aMED (25) and DASH (26) consider red and processed meat consumption unhealthy, while HEI-2015 does not assess those specifically, and aMED further includes alcohol intake and considers moderate drinking healthy. Like our findings, Thomson et al. (11) found that individual components of HEI-2005 were not associated with ovarian cancer survival whereas the total score was, indicating overall dietary quality and/or pattern could play a more crucial role in ovarian cancer survivorship than individual foods. Considering the DASH diet was designed for hypertension prevention, DASH might not be as relevant for gynecological cancer risk or outcomes.

We used a semi-Markov multistate framework to account for death as a semicompeting event for ovarian cancer and Cox models to obtain hazard ratios, making us, as far as we know, the first study to assess both risk and survival in the same population using the same model. If we ignored the semicompeting risk of death and censored individuals based on their death in the Cox models, we would have violated the assumption of independence of the censoring distribution, which would have led to an overestimation of risk and thus biased estimates. Our semicompeting risk model is equivalent to a classical illness-death model (38,39), which is a special case of a multistate model (40). We used the semi-Markov model as opposed to the Markov model because we did not believe we could assume that probability of transitioning to another state was only dependent on the present state (Markov assumption) but that the failure time was dependent on the patient’s history through the present state and the time since entering the present state (semi-Markov assumption). In our study, age (time since entering baseline state) is a risk factor for ovarian cancer, and time since diagnosis (time since entering the ovarian cancer state) is a predictor for survival, thus our model qualifies as a semi-Markov model.

Compared with prior research on diet quality and ovarian cancer risk and survival, our study has the largest sample size and longest follow-up, which allowed us to control for a wide range of confounders. To our knowledge, we are also the first to evaluate risk and survival in the same dataset. We also used standardized dietary scores to ensure comparability with previous publications. Additionally, by using semi-Markov models, we accounted for the semicompeting risk of death ensuring we obtained the most valid estimates for our risk outcome. However, our study is not without limitations. First, diet was only assessed at 1 time point during participants’ mid to late adulthood. Additional research on diet quality at different life stages as well as at or after diagnosis in relation to ovarian cancer is crucial, as well as changes in diet quality over the cancer trajectory. Diet was also assessed via self-report; so poor recall or measurement error could have led to nondifferential exposure misclassification. Finally, our population was largely non-Hispanic White people and highly educated, which may limit the generalizability of our findings.

In summary, we found better diet quality prior to ovarian cancer diagnosis was associated with lower all-cause mortality but found no evidence of an association with ovarian cancer risk. Because our results indicate prediagnosis diet quality is relevant for survival after ovarian cancer diagnosis, more complementary research on postdiagnosis diet quality is needed to understand if addressing diet quality after diagnosis in ovarian cancer survivors could improve outcomes. Although not solely a nutrition intervention, the ongoing Lifestyle Intervention in Ovarian Cancer Enhanced Survival intervention study is assessing improving diet quality postdiagnosis combined with increasing physical activity on progression-free survival among patients with stage II-IV ovarian cancer (41) and will provide critical evidence to inform lifestyle guidelines for ovarian cancer survivorship.

Data availability

Data described in the manuscript, code book, and analytic code will be made available upon request, pending approval from the NIH-AARP Diet and Health Study Steering Committee. Further details are provided at https://www.nihaarpstars.com/.

Author contributions

Anlan Cao, MBBS (Conceptualization; Data curation; Formal analysis; Investigation; Methodology; Writing—original draft), Denise A. Esserman, PhD (Formal analysis; Funding acquisition; Investigation; Methodology; Software; Supervision; Writing—review & editing), Brenda Cartmel, PhD (Formal analysis; Investigation; Methodology; Supervision; Writing—review & editing), Melinda L. Irwin, PhD, MPH (Formal analysis; Funding acquisition; Investigation; Supervision; Writing—review & editing), and Leah M. Ferrucci, PhD, MPH (Conceptualization; Data curation; Formal analysis; Funding acquisition; Investigation; Methodology; Supervision; Writing—review & editing).

Funding

This study was supported by the National Cancer Institute at the National Institutes of Health (NCI P30 CA016359) and by the National Center for Advancing Translational Science at the National Institutes of Health (UL1TR000142, UL1 TR001863) and the Claude D. Pepper Older Americans Independence Center at Yale (P30AG021342).

Conflicts of interest

The authors declare no conflicts of interest.

Acknowledgements

The study sponsor had no role in the design of the study, data collection, the analysis or interpretation of the data, the writing of the article, or the decision to submit for publication.