Dietary Fatty Acid Intake and Prostate Cancer Survival
Dietary Fatty Acid Intake and Prostate Cancer Survival
The study population consisted of 525 men diagnosed with incident prostate cancer and originally recruited for a population-based case-control study in Örebro County, Sweden, during 2 periods: January 1989–September 1991 and May 1992–July 1994 (participation rate, 80.6%). Eligible participants were born in Sweden, aged 80 years or less, and living in Örebro County at the time of diagnosis. All cases of prostate cancer were cytologically and/or histologically confirmed by a study pathologist. As regular prostate cancer screening did not occur in this population at the time of study, most cases were diagnosed because of prostate-related symptoms. The current study was approved by the ethics review board of Uppsala University, Sweden.
Participants completed a self-administered food frequency questionnaire to assess regular diet in the year prior to diagnosis, with the majority of questionnaires completed within 3 months of diagnosis. During the first study period (January 1989–September 1991), nondietary factors were assessed through in-person interviews. Cases recruited during the second study period (May 1992–July 1994) completed mailed questionnaires extended to include questions on nondietary factors; questionnaires were completed by phone when necessary. Body mass index was calculated by using clinical measurements. Information on primary treatment for prostate cancer was obtained through medical records reviewed by trained study staff.
The food frequency questionnaire included 68 food items selected to represent the common Swedish diet during the study period. Frequencies of consumption were multiplied by age-specific standard portion sizes based on the 1988 Swedish National Food Administration handbook and by the nutrient composition of foods specific to Sweden to determine daily energy and nutrient intakes. A validation study was conducted among 87 control subjects enrolled in the original case-control study who completed four 1-week dietary records, given 4 times over the course of 1 year. Pearson's correlation coefficients between energy-adjusted nutrients assessed from the questionnaires and dietary records ranged between 0.2 and 0.6, with r = 0.5 for energy intake, saturated fat, and polyunsaturated fat and r = 0.4 for total fat; specific fatty acids were not reported. Dietary intake of 14 fatty acids was calculated from questionnaire data: saturated lauric (C12:0), myristic (C14:0), palmitic (C16:0), stearic (C18:0), and arachidic (C20:0) acids; monounsaturated palmitoleic (C16:1) and oleic (C18:1) acids; n-3 polyunsaturated alpha-linolenic (C18:3), eicosapentaenoic (EPA; C20:5), docosapentaenoic (DPA; C22:5), and docosahexaenoic (DHA; C22:6) acids; n-6 polyunsaturated linoleic (C18:2) and arachidonic (C20:4) acids; and a composite variable representing shorter chain saturated fatty acids (C4-C10).
Tumors were graded by board-certified pathologists according to the 1980 World Health Organization criteria and staged by using the 1978 TNM classification system. We defined localized tumors as those confined to the prostate at diagnosis (stage T0-T2/M0). Tumors that had progressed through the capsule or metastasized at diagnosis were defined as advanced stage (T3-T4/M0, T0-T4/M1). The study population was routinely screened for skeletal metastases.
The men have been followed prospectively for mortality and cause of death through linkage to the Swedish Cause of Death Registry. In Sweden, individuals are linked to national comprehensive health registries through a personal identification number. In this manner, we achieved complete follow-up of our study population. Cause of death was confirmed through medical record review by a committee of study urologists (S. O. A., O. A., Jan-Erik Johansson).
Dietary intake of each fatty acid was calculated in grams per day (g/day), and values were log10 transformed to improve normality. We calculated age- and energy-adjusted Spearman's correlation coefficients to investigate the correlation among the 14 fatty acids. Fatty acid intake was adjusted for non-alcohol energy intake by the residual method and categorized into quartiles. We also examined the association between individual fatty acid intake and prostate cancer-specific mortality, modeling a 1-standard deviation increment in fatty acid intake continuously.
Cox proportional hazards models were constructed to estimate hazard ratios and 95% confidence intervals, with the lowest quartile of intake for each fatty acid as the referent. Cox models were also utilized to estimate the hazard ratio per standard deviation increment of each individual fatty acid. Follow-up time was calculated from date of cancer diagnosis to death from prostate cancer or death from other causes, or it was censored at the end of follow-up (March 1, 2011).
Our primary analysis focused on time to death from prostate cancer associated with dietary intake of each individual fatty acid, as well as total fat intake. Subsequent analyses grouped fatty acids according to saturation level (saturated, monounsaturated, n-3 polyunsaturated, n-6 polyunsaturated). To assess potential competing risks, we examined time to death from other causes and all-cause mortality. Multivariable models were adjusted for age at diagnosis (<65, 65–69, 70–74, ≥75 years), body mass index (weight (kg)/height (m)2), smoking status (never, former, current), family history of prostate cancer (father or brother), calendar year of diagnosis (1989–1991, 1992–1994), and alcohol intake (nondrinker, <55 g/week, ≥55 g/week). Four men missing data on body mass index were assigned the mean value (25.9 kg/m). We found no evidence of confounding by treatment, and the variable was not retained in final models. Although we have information on tumor cell differentiation (World Health Organization categories: well, moderate, poor), this may be an intermediate on the causal pathway if an association exists and, thus, was not included in final models.
We examined the ratio of n-3 to n-6 polyunsaturated fatty acids, following experimental evidence suggesting opposing effects on tumor growth, and the balance between the 2 groups may affect prostate cancer risk and tumor characteristics. We compared quartiles of total n-3 with total n-6 fatty acid intake and also compared intake of the most common n-3 fatty acid, alpha-linolenic acid, with the most common n-6 fatty acid, linoleic acid.
We tested for linear trend across categorical models by modeling the median of each fatty acid quartile as a semicontinuous variable and including this variable in a multivariable model. All models were additionally stratified by stage at diagnosis (localized or advanced) to assess whether associations with survival varied by clinical stage. We conducted a sensitivity analysis excluding all deaths occurring within the first 2 years of follow-up to minimize the possibility that men with the most severe disease recalled fat intake differently from other men. All tests were 2 sided, and P < 0.05 was considered statistically significant. Analyses were conducted by using SAS, version 9.1, software (SAS Institute, Inc., Cary, North Carolina).
The proportional hazards assumption was tested by adding an interaction term between the median value for each fatty acid quartile and follow-up time (continuous) to the multivariable model of the fatty acid main effect (quartiles). The proportional hazards assumption was satisfied for 12 fatty acids; however, significant interaction terms in the models of palmitoleic and stearic acids suggest that the association between those fatty acids and survival may vary with time. This discrepancy should be noted when interpreting models of the 2 acids.
We explored an interaction between stage at diagnosis and total fat intake with multivariable Cox models that included an indicator variable for advanced stage and the product term of stage with fat as a continuous variable. A likelihood ratio test with 1 df assessed statistical significance, comparing the above model with one without the product term.
We further adjusted models of marine-derived fatty acids (EPA, DPA, DHA) for energy-adjusted vitamin D intake (continuous) because fish is a source of vitamin D, and vitamin D may be associated with improved prostate cancer survival. We also constructed a summary variable combining dietary EPA, DPA, and DHA. The new variable was assessed per standard deviation increment and per energy-adjusted quartiles to explore associations with prostate cancer-specific mortality overall and by stage at diagnosis.
Because patterns of fatty acid intake rather than specific fatty acids may be important predictors of prostate cancer outcome, we conducted a principal components analysis to identify a smaller set of variables that explained the majority of the total variance of the 14 original fatty acids. Principal components, or eigenvectors, were retained until 90% of the total variance could be explained, following the percentage of variance criterion. Three eigenvectors were included in a multivariable model to explore an association with prostate cancer-specific survival, overall and stratified by stage at diagnosis.
Materials and Methods
Study Population
The study population consisted of 525 men diagnosed with incident prostate cancer and originally recruited for a population-based case-control study in Örebro County, Sweden, during 2 periods: January 1989–September 1991 and May 1992–July 1994 (participation rate, 80.6%). Eligible participants were born in Sweden, aged 80 years or less, and living in Örebro County at the time of diagnosis. All cases of prostate cancer were cytologically and/or histologically confirmed by a study pathologist. As regular prostate cancer screening did not occur in this population at the time of study, most cases were diagnosed because of prostate-related symptoms. The current study was approved by the ethics review board of Uppsala University, Sweden.
Exposure Assessment
Participants completed a self-administered food frequency questionnaire to assess regular diet in the year prior to diagnosis, with the majority of questionnaires completed within 3 months of diagnosis. During the first study period (January 1989–September 1991), nondietary factors were assessed through in-person interviews. Cases recruited during the second study period (May 1992–July 1994) completed mailed questionnaires extended to include questions on nondietary factors; questionnaires were completed by phone when necessary. Body mass index was calculated by using clinical measurements. Information on primary treatment for prostate cancer was obtained through medical records reviewed by trained study staff.
The food frequency questionnaire included 68 food items selected to represent the common Swedish diet during the study period. Frequencies of consumption were multiplied by age-specific standard portion sizes based on the 1988 Swedish National Food Administration handbook and by the nutrient composition of foods specific to Sweden to determine daily energy and nutrient intakes. A validation study was conducted among 87 control subjects enrolled in the original case-control study who completed four 1-week dietary records, given 4 times over the course of 1 year. Pearson's correlation coefficients between energy-adjusted nutrients assessed from the questionnaires and dietary records ranged between 0.2 and 0.6, with r = 0.5 for energy intake, saturated fat, and polyunsaturated fat and r = 0.4 for total fat; specific fatty acids were not reported. Dietary intake of 14 fatty acids was calculated from questionnaire data: saturated lauric (C12:0), myristic (C14:0), palmitic (C16:0), stearic (C18:0), and arachidic (C20:0) acids; monounsaturated palmitoleic (C16:1) and oleic (C18:1) acids; n-3 polyunsaturated alpha-linolenic (C18:3), eicosapentaenoic (EPA; C20:5), docosapentaenoic (DPA; C22:5), and docosahexaenoic (DHA; C22:6) acids; n-6 polyunsaturated linoleic (C18:2) and arachidonic (C20:4) acids; and a composite variable representing shorter chain saturated fatty acids (C4-C10).
Outcome Ascertainment
Tumors were graded by board-certified pathologists according to the 1980 World Health Organization criteria and staged by using the 1978 TNM classification system. We defined localized tumors as those confined to the prostate at diagnosis (stage T0-T2/M0). Tumors that had progressed through the capsule or metastasized at diagnosis were defined as advanced stage (T3-T4/M0, T0-T4/M1). The study population was routinely screened for skeletal metastases.
The men have been followed prospectively for mortality and cause of death through linkage to the Swedish Cause of Death Registry. In Sweden, individuals are linked to national comprehensive health registries through a personal identification number. In this manner, we achieved complete follow-up of our study population. Cause of death was confirmed through medical record review by a committee of study urologists (S. O. A., O. A., Jan-Erik Johansson).
Statistical Analysis
Dietary intake of each fatty acid was calculated in grams per day (g/day), and values were log10 transformed to improve normality. We calculated age- and energy-adjusted Spearman's correlation coefficients to investigate the correlation among the 14 fatty acids. Fatty acid intake was adjusted for non-alcohol energy intake by the residual method and categorized into quartiles. We also examined the association between individual fatty acid intake and prostate cancer-specific mortality, modeling a 1-standard deviation increment in fatty acid intake continuously.
Cox proportional hazards models were constructed to estimate hazard ratios and 95% confidence intervals, with the lowest quartile of intake for each fatty acid as the referent. Cox models were also utilized to estimate the hazard ratio per standard deviation increment of each individual fatty acid. Follow-up time was calculated from date of cancer diagnosis to death from prostate cancer or death from other causes, or it was censored at the end of follow-up (March 1, 2011).
Our primary analysis focused on time to death from prostate cancer associated with dietary intake of each individual fatty acid, as well as total fat intake. Subsequent analyses grouped fatty acids according to saturation level (saturated, monounsaturated, n-3 polyunsaturated, n-6 polyunsaturated). To assess potential competing risks, we examined time to death from other causes and all-cause mortality. Multivariable models were adjusted for age at diagnosis (<65, 65–69, 70–74, ≥75 years), body mass index (weight (kg)/height (m)2), smoking status (never, former, current), family history of prostate cancer (father or brother), calendar year of diagnosis (1989–1991, 1992–1994), and alcohol intake (nondrinker, <55 g/week, ≥55 g/week). Four men missing data on body mass index were assigned the mean value (25.9 kg/m). We found no evidence of confounding by treatment, and the variable was not retained in final models. Although we have information on tumor cell differentiation (World Health Organization categories: well, moderate, poor), this may be an intermediate on the causal pathway if an association exists and, thus, was not included in final models.
We examined the ratio of n-3 to n-6 polyunsaturated fatty acids, following experimental evidence suggesting opposing effects on tumor growth, and the balance between the 2 groups may affect prostate cancer risk and tumor characteristics. We compared quartiles of total n-3 with total n-6 fatty acid intake and also compared intake of the most common n-3 fatty acid, alpha-linolenic acid, with the most common n-6 fatty acid, linoleic acid.
We tested for linear trend across categorical models by modeling the median of each fatty acid quartile as a semicontinuous variable and including this variable in a multivariable model. All models were additionally stratified by stage at diagnosis (localized or advanced) to assess whether associations with survival varied by clinical stage. We conducted a sensitivity analysis excluding all deaths occurring within the first 2 years of follow-up to minimize the possibility that men with the most severe disease recalled fat intake differently from other men. All tests were 2 sided, and P < 0.05 was considered statistically significant. Analyses were conducted by using SAS, version 9.1, software (SAS Institute, Inc., Cary, North Carolina).
The proportional hazards assumption was tested by adding an interaction term between the median value for each fatty acid quartile and follow-up time (continuous) to the multivariable model of the fatty acid main effect (quartiles). The proportional hazards assumption was satisfied for 12 fatty acids; however, significant interaction terms in the models of palmitoleic and stearic acids suggest that the association between those fatty acids and survival may vary with time. This discrepancy should be noted when interpreting models of the 2 acids.
We explored an interaction between stage at diagnosis and total fat intake with multivariable Cox models that included an indicator variable for advanced stage and the product term of stage with fat as a continuous variable. A likelihood ratio test with 1 df assessed statistical significance, comparing the above model with one without the product term.
We further adjusted models of marine-derived fatty acids (EPA, DPA, DHA) for energy-adjusted vitamin D intake (continuous) because fish is a source of vitamin D, and vitamin D may be associated with improved prostate cancer survival. We also constructed a summary variable combining dietary EPA, DPA, and DHA. The new variable was assessed per standard deviation increment and per energy-adjusted quartiles to explore associations with prostate cancer-specific mortality overall and by stage at diagnosis.
Because patterns of fatty acid intake rather than specific fatty acids may be important predictors of prostate cancer outcome, we conducted a principal components analysis to identify a smaller set of variables that explained the majority of the total variance of the 14 original fatty acids. Principal components, or eigenvectors, were retained until 90% of the total variance could be explained, following the percentage of variance criterion. Three eigenvectors were included in a multivariable model to explore an association with prostate cancer-specific survival, overall and stratified by stage at diagnosis.
Source...