• ovarian cancer;
  • diet;
  • survival


  1. Top of page
  2. Abstract
  6. Acknowledgements

We evaluated the effects of various food groups and micronutrients in the diet on survival among women who originally participated in a population-based case-control study of ovarian cancer conducted across 3 Australian states between 1990 and 1993. This analysis included 609 women with invasive epithelial ovarian cancer, primarily because there was negligible mortality in women with borderline tumors. The women's usual diet was assessed using a validated food frequency questionnaire. Deaths in the cohort were identified using state-based cancer registries and the Australian National Death Index (NDI). Crude 5-year survival probabilities were estimated using the Kaplan-Meier technique, and adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) were obtained from Cox regression models. After adjusting for important confounding factors, a survival advantage was observed for those who reported higher intake of vegetables in general (HR = 0.75, 95% CI = 0.57–0.99, p-value trend 0.01 for the highest third, compared to the lowest third), and cruciferous vegetables in particular (HR = 0.75, 95% CI = 0.57–0.98, p-value trend 0.03), and among women in the upper third of intake of vitamin E (HR = 0.76, 95% CI = 0.58–1.01, p-value trend 0.04). Inverse associations were also seen with protein (p-value trend 0.09), red meat (p-value trend 0.06) and white meat (p-value trend 0.07), and modest positive trends (maximum 30% excess) with lactose (p-value trend 0.04), calcium and dairy products. Although much remains to be learned about the influence of nutritional factors after a diagnosis of ovarian cancer, our study suggests the possibility that a diet high in vegetable intake may help improve survival. © 2003 Wiley-Liss, Inc.

Ovarian cancer is a major cause of cancer mortality in women, primarily due to the insidious onset and consequential late diagnosis of the disease in many women. Given the high proportion of late-stage disease at diagnosis, treatment outcomes are often poor, even in specialized gynecologic oncology centers.1 For the time being, advancing diagnosis through screening is infeasible as a serious public health option.2 Similarly, practical strategies for primary prevention remain limited. Efforts to identify other preventive factors or modifiable causes, dietary3, 4 and otherwise,5 are under way, but as yet have not yielded definitive results.

We thus considered it worthwhile to examine whether any lifestyle factors over which women might exert some control could improve their prognosis. We have followed up on women who were enrolled in an etiologic study of ovarian cancer in the early 1990s6 and report here our initial findings regarding possible influences of diet on survival. We are unaware of any previous report of this, or any other nonclinical prognostic factor, for ovarian cancer.


  1. Top of page
  2. Abstract
  6. Acknowledgements

Our study assesses the mortality experience of a large group of Australian women diagnosed with epithelial ovarian cancer in the early 1990s, enrolled in a case-control study,6 and in particular explores the possible influence of various dietary factors on survival. Ethics approval for the research was received from the University of Queensland and all metropolitan hospitals where patients were recruited. All participants gave informed consent prior to enrollment, including granting the study investigators full access to their medical records.

The study population comprised 822 women with incident epithelial ovarian cancer (90% of all eligible cases ascertained) who were diagnosed between 1990 and 1993, aged 18–79 years and identified through major gynecology-oncology centres in the Australian states of Queensland, New South Wales and Victoria. Comparisons with registry data indicated that close to 100% of all Queensland cases, and about half of those diagnosed in the other states, were enrolled. Of a total of 1,116 cases of epithelial ovarian cancer identified, 184 were ineligible, and 19 were unable to complete the questionnaire because they were too ill. Of the remaining 913 eligible cases, 50 died before an interview could take place, 34 refused to participate and for 7 cases the doctor did not grant permission for interview. Women with borderline tumors (n = 145) were excluded, as were 4 women whose dietary data were implausible and another 64 who did not return their dietary questionnaire, or for whom dietary information was incomplete, leaving 609 women for these analyses.

The cohort was followed for mortality for between 5 and 8.3 years post-diagnosis (average 7.3 years), to 30 June 1999. Personal identifiers were used to link the cohort to state cancer registry records and the Australian National Death Index (NDI). Both the NDI and cancer registries used probabilistic record linkage software (Automatch) that identified likely matches based on key identifiers (full name, date of birth, residence address, hospital and diagnosis date).7 To determine the likelihood of a true match, we grouped the records together on the basis of present surname and then clerically reviewed all potential matches from the multiple data sources. We estimate mortality follow-up to be complete. To identify all ovarian cancer-related deaths (and to assist with follow-up), we obtained full details, including immediate cause, underlying cause and contributing factors directly from death certificates

Prognostic variables were ascertained in a variety of ways. Clinical and pathologic information was collected from medical records, and histologic diagnoses were reviewed centrally by a single pathologist in each state. The dietary data were gathered using a semiquantitative food frequency questionnaire (119 food items), with women being asked to report their usual intake over the year prior to the onset of any symptoms they related to their diagnosis. This was self-completed, usually within 2–3 months of diagnosis, and was derived from an early version of the Nurses' Health Study questionnaire,8 modified to reflect Australian foods and portion sizes.9 Height, weight and other personal information were collected in a standardized face-to-face interview as described elsewhere.6

Dietary analyses are challenging and may range from deep exploration of a single food or nutrient or food/nutrient group to a comprehensive, thorough review of all nutrients estimated. We have elected to examine a subset of foods and nutrients, primarily based on those with known or suggested preventive or causal roles with respect to ovarian or other hormone-related cancers, as well as a few that have been identified as possibly influencing breast cancer prognosis. Thus, we have focused on the possible protective roles of fruit, vegetables/nutrients10, 11, 12, 13 and protein/white meat,14 and the putative harmful roles of fat,10, 11, 12, 13, 14, 15 dairy products13, 14, 15 and alcohol.16, 17 Body mass index (BMI) (kg/m2) was also considered in the present analyses due to its close relationship with total energy intake, as well as its possible influence on ovarian carcinogensis5 and breast cancer prognosis.18, 19, 20, 21, 22

Survival time was calculated from the date of diagnosis to the date of death (from ovarian cancer) or censored at the earliest of 30 June 1999 or death from another cause. The Kaplan-Meier technique was used to plot crude survival curves and estimate crude overall survival probabilities, and adjusted hazard ratios (HRs) and 95% confidence intervals (CIs) were obtained from Cox regression models. The p-value for linear trend was calculated by the change in the likelihood ratio statistic for entry of a linear term for the nutrient in the model, and thus was a chi-square test on 1 degree of freedom. All dietary analyses were adjusted for age (10-year age groups), tumor stage, tumor grade, total energy intake (thirds) and BMI (thirds). Nutrients were entered as their energy-adjusted residuals.23 Dietary analyses included only single dietary factors in the core model, except where otherwise indicated, and dietary exposures were split at their tertiles. All analyses were performed using the Statistical Packages for Social Sciences for Windows, version 10.0 (SPSS Inc., Chicago, IL).


  1. Top of page
  2. Abstract
  6. Acknowledgements

There was a marked difference in crude survival according to tumor invasiveness, with only 2 of 145 women with borderline tumors dying during the follow-up period (1%). Consequently, these women were omitted from all subsequent analyses. Among the 609 women with invasive epithelial ovarian cancer and complete dietary questionnaires, 372 (61%) died from ovarian cancer during follow-up, and 22 from other causes. The 5-year survival for these women was 45% (standard error [SE] = 2%). Our analyses refer only to ovarian cancer mortality; competing causes were censored.

Increasing tumor stage, histologic grade and patient age at diagnosis had clear adverse effects on survival (Table I). These 3 factors were therefore included in all subsequent models (tumor grades moderate, poor and undifferentiated were combined into a single group) in addition to total energy intake (kilocalories) and BMI. Further adjustment for other variables known to influence survival in this cohort (e.g., residual disease and ascites) or for other known risk factors for ovarian cancer (e.g., parity, length of oral contraceptive pill use and smoking) made no difference to the estimates of effect.

Table I. The effect of principle non-nutrient variables on ovarian cancer survival
VariableTotal% DeadCrude 5-year survival % (SE %)Adjusted1 hazard ratio (95% CI)p trend2
  • 1

    Adjusted for FIGO stage, age, grade, total energy intake (Kilocalories), BMI, residual, ascites, smoking status, parity and length of OCP use.

  • 2

    Missing and not recorded categories excluded from p trend test.

  • 3

    Federation of Gynecology and Obstetrics.

  • 4

    Histopathologic grade.

  • 5

    Unable to locate medical record.

  • 6

    Length of use of oral contraceptive pill.

  • 7

    BMI prior to illness.

FIGO stage3     
 I1441888 (3)1.0 
 II583865 (6)2.30 (1.28–4.12) 
 III3387827 (2)5.65 (3.64–8.76) 
 IV589312 (4)8.19 (4.79–14.01)<0.001
Age group     
 <40 years483567 (7)1.0 
 40–49 years1195158 (5)1.28 (0.74–2.23) 
 50–59 years1766148 (4)1.57 (0.92–2.68) 
 60–69 years1696836 (4)1.97 (1.15–3.37) 
 70–79 years977429 (5)1.99 (1.14–3.49)<0.001
 Well differentiated872579 (4)1.0 
 Moderately differentiated1596441 (4)2.24 (1.42–3.85) 
 Poorly differentiated3346939 (3)2.35 (1.45–3.79) 
 Undifferentiated197430 (11)3.11 (1.48–6.53)<0.001
Residual disease     
 None–<1cm2445056 (3)1.0 
 1–2cms688421 (5)1.20 (0.85–1.69) 
 >2cms898913 (4)1.50 (1.09–2.05) 
 Not recorded1245951 (5)1.17 (0.84–1.62) 
 Missing record584   0.01
 None1444262 (4)1.0 
 <500 mls356045 (8)0.98 (0.58–1.65) 
 >500 mls1848223 (3)1.70 (1.23–2.36) 
 Not recorded1626147 (4)1.23 (0.85–1.76) 
 Missing record584   0.01
Smoking status     
 Never smoked3666246 (3)1.0 
 Current smoker1085250 (5)1.53 (1.11–2.12) 
 Ex-smoker1356639 (4)1.17 (0.90–1.53)0.12
 01355055 (4)1.0 
 1826641 (5)1.27 (0.86–1.87) 
 21786642 (4)1.02 (0.74–1.40) 
 31165650 (5)0.71 (0.49–1.02) 
 4556638 (7)1.12 (0.73–1.74) 
 5437235 (7)0.93 (0.58–1.49)0.30
Length of use of OCP6     
 Never used3226738 (3)1.0 
 <24 months915949 (5)1.08 (0.77–1.52) 
 24–59 months754756 (6)0.62 (0.42–0.94) 
 60–119 months685357 (6)0.92 (0.62–1.37) 
 >120 months525851 (7)1.03 (0.67–1.59)0.50
Usual BMI7 (kg/m2)     
 <22.21985945 (4)1.0 
 22.2–25.82066343 (3)1.01 (0.78–1.31) 
 >25.82056247 (4)0.96 (0.74–1.23)0.43

When divided into thirds, usual BMI was unrelated to longevity (Table I). However, the very slimmest women (BMI < 20 kg/m2) did have somewhat better survival than the rest (HR = 0.8, 95% CI = 0.48–1.35).

Results of analyses focusing on hypothesized protective factors, mostly vegetables, fruits and related micronutrients, are shown in Table II. Most notable were the inverse associations with increasing intake of vegetables overall (Fig. 1), and of cruciferous vegetables, and with vitamin E from foods and meats. There was a suggestion of an inverse association for higher intake of fruit, but there was no discernable influence of vitamin C (from foods), fibre or tea (not shown). The absolute adjusted survival advantages at 5 years were 13% for the highest third of vegetable intake (median of 6.9 servings daily) compared to the third eating least vegetables (median 3.1 servings daily) and 10% for women in the highest vs. lowest third of vitamin E intake from foods. There was a suggestion of a modest inverse trend with protein, red and white meat (fish and poultry).

Table II. The effects on various food groups and components on ovarian cancer survival
Daily intakeTotal% DeadCrude 5-year survival % (SE %)Adjusted1 hazard ratio (95% CI)p trend
  • 1

    Adjusted for FIGO stage, age, grade, total energy intake (Kilocalories), BMI.

  • 2

    All vegetables without potato.

  • 3

    Broccoli, cauliflower, cabbage, coleslaw, brussels-sprouts.

  • 4

    Carrot, sweet potato, pumpkin.

  • 5

    Categories based on distribution of nutrient residuals, the “raw” median value for each category is shown in brackets.

  • 6

    Poultry and fish.

All vegetables2     
 <3.9 serves2016444 (4)1.0 
 3.9–5.56 serves2016343 (4)1.08 (0.82–1.42) 
 >5.56 serves2075850 (3)0.75 (0.57–0.99)0.01
Cruciferous vegetables3     
 <0.41 serves2016542 (4)1.0 
 0.41–0.83 serves2036144 (4)0.87 (0.67–1.13) 
 >0.83 serves2055849 (4)0.75 (0.57–0.98)0.03
Yellow vegetables4     
 <0.6 serves1826244 (4)1.0 
 0.6–0.99 serves1965947 (4)0.96 (0.74–1.25) 
 >0.99 serves2316245 (3)0.98 (0.73–1.31)0.72
All fruit     
 <2.79 serves2025946 (4)1.0 
 2.79–4.49 serves2006145 (4)0.95 (0.72–1.26) 
 >4.49 serves2076344 (3)0.89 (0.67–1.18)0.59
 Low [18.2g]1885949 (4)1.0 
 Medium [26 g]2135847 (3)1.02 (0.77–1.34) 
 High [35.5 g]2086740 (3)0.99 (0.75–1.29)0.99
 <192 g2016542 (4)1.0 
 192–263 g2065848 (3)0.73 (0.53–0.99) 
 >263 g2026044 (3)0.85 (0.57–1.28)0.48
Vitamin E (from foods)5     
 Low [7.2 mg]2046344 (4)1.0 
 Medium [10.1 mg]2096144 (4)0.94 (0.72–1.22) 
 High [12.4 mg]1966048 (4)0.76 (0.58–1.01)0.04
Vitamin C (from foods)5     
 Low [108 mg]1956046 (4)1.0 
 Medium [170 mg]2066146 (4)1.01 (0.77–1.31) 
 High [256 mg]2085044 (3)0.92 (0.71–1.21)0.65
 <4387 μg2016147 (4)1.0 
 4387–6930 μg2016144 (4)1.09 (0.83–1.43) 
 >6930 μg2076145 (3)1.08 (0.82–1.42)0.85
 <74.5 g2056540 (3)1.0 
 74.5–98.5 g2026048 (4)0.80 (0.60–1.07) 
 >98.5 g2025946 (3)0.72 (0.50–1.04)0.09
White Meat6     
 <0.3 serves2026540 (3)1.0 
 0.3–0.54 serves1915848 (4)0.81 (0.63–1.06) 
 >0.54 serves2166147 (3)0.78 (0.6–1.01)0.07
Red meat     
 <0.5 serves2096740 (3)1.0 
 0.5–0.86 serves1895847 (4)0.89 (0.68–1.18) 
 >0.86 serves2115849 (4)0.76 (0.58–1.00)0.06
thumbnail image

Figure 1. Kaplan-Meier survival curves for vegetable consumption. Low, <3.9 servings/day; Medium, 3.9–5.6 servings/day; High, >5.6 servings/day.

Download figure to PowerPoint

Very few women in the cohort had taken vitamins or supplements. Only 58 women reported taking vitamin E from supplements, and they did not appear to have any survival advantage, nor did women who reported taking beta-carotene or vitamin C supplements or multivitamins

Macronutrients and dietary factors thought a priori to be more likely to have a negative influence on prognosis are included in Table III. Of most note here were the positive associations, and hence worse survival, seen with increasing intake of lactose, dairy products and calcium, with the highest third of intake carrying about a 30% excess risk of early death compared to the lowest third. Of the principle energy sources, fats and alcohol (not shown) had little influence on outcome, but there was a suggestion of a modest inverse trend with carbohydrate consumption. Survival was relatively independent of energy intake per se.

Table III. Possible a priori “harmful” prognostic factors for ovarian cancer
Daily intakeTotal% DeadCrude 5-year survival % (SE %)Adjusted1 hazard ratio (95% CI)p trend
  • 1

    Adjusted for stage, age, grade, total energy intake (Kilocalories), BMI.

  • 2

    Categories based on distribution of nutrient residuals, the “raw” median value for each category is shown in brackets.

Dairy products     
 <1.86 serves2066046 (4)1.0 
 1.86–3.16 serves2045848 (4)0.97 (0.74–1.27) 
 >3.16 serves1996541 (4)1.30 (0.97–1.74)0.11
 <13.6 g1996248 (4)1.0 
 13.6–20.17 g2045847 (4)0.99 (0.75–1.31) 
 >20.17 g2066640 (3)1.32 (0.99–1.75)0.04
 Low [610 mg]2015452 (4)1.0 
 Medium [884 mg]2076641 (3)0.99 (0.75–1.29) 
 High [1285 mg]2016343 (4)1.27 (0.97–1.67)0.08
Fat total2     
 Low [58 g]2066145 (4)1.0 
 Medium [76 g]2216443 (3)1.09 (0.84–1.40) 
 High [86 g]1825848 (4)1.06 (0.81–1.39)0.42
Saturated fat2     
 Low [22 g]2066245 (3)1.0 
 Medium [28 g]2135946 (3)1.06 (0.82–1.36) 
 High [37 g]1906344 (4)1.13 (0.87–1.47)0.25
Polyunsaturated fat2     
 Low [7 g]2015946 (4)1.0 
 Medium [11 g]2116642 (3)0.99 (0.76–1.29) 
 High [15 g]1975848 (4)0.91 (0.69–1.20)0.87
Monounsaturated fat2     
 Low [18 g]2076045 (4)1.0 
 Medium [25 g]2186839 (3)1.10 (0.85–1.41) 
 High [30 g]1845353 (4)0.91 (0.68–1.21)0.86
 <17071996344 (4)1.0 
 1707–22122036443 (3)0.91 (0.71–1.18) 
 >22122075748 (4)0.92 (0.70–1.19)0.62

We considered simple joint effect models to assess confounding by the more influential dietary factors (both positive and negative) and found modest intercorrelation between the effects of lactose and calcium, but otherwise the HRs were little affected.


  1. Top of page
  2. Abstract
  6. Acknowledgements

Overall survival from ovarian cancer was poor in this cohort, with only 45% of women alive 5 years after diagnosis. Nonetheless, this preliminary exploration of the influence of dietary factors raises the possibility that small but potentially valuable improvements in survival may be possible through modest increases in vegetable intake, perhaps the cruciferous vegetables in particular. A similar inverse association was seen among women in the upper third of intake of vitamin E from foods, and for red and white meat (and, perhaps, protein overall). The only other relation with lower mortality was among lean women whose usual prediagnostic BMI was less than 20. On the harmful side of the dietary equation, only the constellation of dairy-related factors showed even a modest relation with poorer survival.

The women in this cohort were unselected with regard to either prognosis or diet, and follow-up was essentially complete. Both assignment of cause of death and baseline clinical measures were made independently of diet, and are likely to have been quite accurate. Dietary assessment was unbiased with respect to outcome, but will have been subject to random error. While this may have hidden real associations, it is unlikely to be responsible for apparent variations in survival across these intake groups.

A potential limitation of the study is the possibility of selection bias introduced by eligible cases that did not participate due to illness or death (n = 69) and those who did not return their dietary questionnaires (n = 64). Although this is a large case-control study, it was not a true population series, and dietary data were collected after diagnosis. Women who did not return their dietary questionnaire were older (mean age of 60 years, compared to 57 years) and had more stage 3 and 4 disease (80% compared to 65%). Clearly those eligible cases who did not return their dietary questionnaires were a poor prognostic group, but it seems unlikely that their absence contributed to the dietary associations we observed. Indeed, if our findings are reliable, then we could speculate that this group may well have a less optimal diet than those who died later; i.e., that our findings would have been biased downward by their loss. Our study sample therefore underrepresents women with progressive, terminal disease, and thus is a more favorable prognostic group than the population.

The apparently influential dietary elements could simply be chance findings that are to be expected among the many hundreds of potential associations we could have studied (but to date have not). The simple joint effect models indicated modest intercorrelation between lactose and calcium, but otherwise HRs were little affected when other dietary factors were added to the models. However, the effects seen here could still be markers for other unstudied dietary elements, including more complex dietary factors, related personal or lifestyle attributes or postdiagnostic changes in any of these aspects.

The present analyses are based on a woman's reported usual diet pre-diagnosis. We have no clear biologic basis for weighing the importance of usual pre-diagnostic diet and what a woman eats after having her cancer diagnosed and treated. It seems probable that post-diagnostic diet could be the more influential, and this is certainly the only aspect that a woman can change. If women did change their diet after diagnosis, the apparently beneficial effects we observed may reflect a propensity for change to a more helpful diet, rather than absolute benefits of previously eating vegetables per se. It is, however, conceivable that despite having failed as a primary preventive, continuing to eat a healthier diet may yet yield an independent contribution to survival. There are some data on diet and breast cancer survival from the Nurses' Health Study that bear on this issue.14 In their study of almost 2,000 women with breast cancer, Holmes et al.14 found that correlations of foods eaten before and after diagnosis were between 0.4 and 0.6, and that the effects of prediagnostic intake of a factor tended to be qualitatively similar, albeit somewhat attenuated from that seen for postdiagnostic diet. This could be taken to suggest that prediagnostic and postdiagnostic diet are relevant to survival, but it does not address directly the importance of specific dietary changes.

As the first study of these associations of which we are aware, there is no direct epidemiologic context in which to embed our findings. However epidemiologic and clinical evidence from the breast cancer literature seems to indicate that several nutritional factors modify the progression and prognosis of breast cancer. In particular, being overweight or obese is associated with a poorer prognosis and is thought to be related to either the effect of excess adipose tissue on circulating gonadal hormones, the interaction between insulin and insulin-like growth factor 1 with adiposity and weight gain or possibly chemotherapy dose reductions that may be occurring in obese women.10 The results of studies that have examined the relationship between dietary fat intake and survival from breast cancer are conflicting, as are those between alcohol intake and prognosis.10, 14, 16, 17 However, the majority of studies that have examined vegetable and/or fruit intakes (or related micronutrients) have reported modest inverse associations.10 Two large multicenter intervention trials are currently under way to examine if diet modification can directly influence the risk for recurrence and survival after a diagnosis of breast cancer.24 The results of these trials will be available within the next 3–4 years.

Many substances in vegetables and fruit have been shown or postulated to have anticarcinogenic properties, and this is supported quite well by the epidemiologic literature. Cruciferous vegetables, for example, contain high levels of dithiolthiones and isothiocyanates, which have been shown to increase the activity of enzymes involved in the detoxification of carcinogens.25 Cruciferous vegetables also contain indole-3-carbinol, which possibly results in the production of a less potent form of estradiol, which may protect against estrogen-related cancers such as those of the breast, endometrium and ovary.12 The role of vitamin E is perhaps less clear, but may involve its effect as an intracellular antioxidant.25 Other possible mechanisms for the protective effect of vegetables and fruit include effects on cell differentiation, increased activity of enzymes that detoxify carcinogens and the preservation of integrity of intracellular matrices, with the latter being perhaps particularly relevant in antagonizing spread of metastases.25

With respect to the positive association with lactose and dairy foods seen here, the proposed biologic mechanism for an etiologic role of lactose and galactose may be relevant,26 although the issue remains unresolved epidemiologically. Our own assessment in these women was that these factors had no etiologic effect,3 so the finding of a possible adverse influence prognostically was somewhat surprising.

Despite the various uncertainties, and given that the relations observed here are truly independent of treatment and clinico-pathologic predictors, these findings give some hope that women with this often fatal condition may be able to influence their fate to a worthwhile degree. In addition to expanded prospective research as described above, this testing should include more comprehensive exploration of dietary associations than attempted here, including modeling dietary patterns, both as risk factors in themselves and for fuller confounder control of the effects of single food groups, foods and nutrients.27 Our results certainly warrant more critical testing and should not be taken as more than suggestive until such tests have been applied.


  1. Top of page
  2. Abstract
  6. Acknowledgements

Initial funding to collect the cases for the survey of Women's Health was obtained from National Health & Medical Research Council and Queensland Cancer Fund.


  1. Top of page
  2. Abstract
  6. Acknowledgements