Source of study population was from the National Health and Nutrition Examination Survey (NHANES), a public database that monitors human health conditions and nutrition across ethnic groups in the United States through continuous updates. NHANES obtained resident samples of all races in the United States through multi-stage stratified sampling, including people of all ages from all regions of the country. Therefore, NHANES data is a representative reflection of the health status of United States residents. Researchers can obtain NHANES data for free from the official website (Supplementary Table S1). The Ethics Review Board of the National Center for Health Statistics had approved the NHANES study. All participants provided written informed consent, allowing researchers to use the data as long as the data source is identified in the findings. The NHANES database included 11,183 adult participants from diverse ethnic and geographic backgrounds, between 2003 and 2006. Hypertension was identified through a computer-assisted personal interview system and a questionnaire. 3,390 participants were identified as having hypertension, while 908 participants were excluded due to not completing the Food Frequency Questionnaire (FFQ). Two participants lacked sufficient information to link to the National Death Index (NDI) data, resulting in missing follow-up survival status results and their exclusion from the study. Ultimately, 2,480 participants were enrolled in the study, as illustrated in Figure 1.

The Food Frequency Questionnaire (FFQ) is a tool used to collect information about food intake frequency in the past 12 months, including fruit consumption habits. Participants in this study were asked about their average intake of apples over the last year by responding to the prompt: “How often did you eat apples/ bananas/ pears/pineapples/grapes?” Their eating habits were then analyzed. Fruit consumption data was obtained via an FFQ, which categorized intake levels as follows: unknow, less than once /month, 1–3 times/month, 1–2 times/ week, three to six times/ week, or once or more /day for apples, pears, bananas, pineapples, and grapes.

All-cause mortality was examined using data from the National Death Index (NDI), which recorded death information for study participants from the survey’s commencement through December 31, 2019. The NDI, published by the National Center for Health Statistics, a United States government agency, connects NHANES data with mortality records. Data perturbation techniques were utilized in the NDI to reduce the risk of participant re-identification and maintain patient confidentiality. Study period concludes 10 years after enrollment. For deceased patients, the follow-up duration since their interview was computed based on the quarter and year of their passing. For living patients, follow-up duration was determined using the end of the follow-up period. All NDI data used in this study were obtained from publicly released documents on the website (Supplementary Table S1).

The collection of baseline patient data was facilitated by a computer-assisted personal interview system and a questionnaire survey. This comprehensive dataset encompassed various demographic factors such as gender, age, race, education level, ratio of family income to poverty, as well as health-related information including smoking and alcohol consumption, presence of hypercholesterolemia, diabetes, cardiovascular disease, stroke, lung disease, liver disease, and failing kidneys. Ethnicity was determined by categorizing patients as Mexican Americans, non-Hispanic Blacks, non-Hispanic whites, other Hispanics, or others. A smoking history of at least 100 cigarettes smoked throughout one’s lifetime was used as the criterion for identifying individuals as smokers. A drinker was operationally defined as an individual who had consumed a minimum of 12 cups alcoholic beverages within the preceding 12-month period. The presence of hypertension, hypercholesterolemia, diabetes, stroke, liver disease, and renal failure was ascertained based on prior diagnoses made by the participant’s healthcare provider. Participants who responded affirmatively to any of the following inquiries were categorized as having cardiovascular disease: “Have you ever been diagnosed with congestive heart failure, coronary heart disease, angina pectoris, or a heart attack by a healthcare professional?”

The median (first quartile, third quartile) was employed for continuous variables that did not adhere to a normal distribution. Categorical variables were expressed as numbers (percentages). The comparison of non-normally distributed continuous variables between two groups was conducted using the rank sum test. Furthermore, Chi-square tests were utilized to compare categorical variables between two groups. Cox proportional hazard model was utilized to examine the impact of fruit intake on all-cause mortality. Model 1 did not adjust any covariates. Model 2 made adjustments for gender, age, race, education level, and ratio of family income to poverty rate. In Model 3, covariates were selected based on the principle of covariate screening: confounding factors were compared before and after adding them to the regression model, and those resulting in a change in p value greater than 10% were included. Adjusting factors in the final Model 3 included sex, age, race, education level, ratio of family income to poverty rate, smoking history, history of hypercholesterolemia, diabetes, history of cardiovascular disease, stroke, lung disease, and history of failing kidneys. Before COX regression, collinearity screening was carried out for all covariates. The results indicated that the Variance Inflation Factor (VIF) for all covariates was less than 5. Splines smoothing plots were utilized to examine the linear or nonlinear correlation between the frequency of consuming different kinds of fruits and all-cause mortality. Additionally, stratified analyses were performed to confirm the consistency of the outcomes in the presence of different diseases. In addition, further elucidating the relationship between fruit intake and mortality was done by using a three-dimensional histogram and a time-mortality curve. The analysis of data was conducted using the R language package and EmPower software.¹ p-values less than 0.05 are considered statistically significant.

A total of 2,480 participants met the study’s exclusion criteria and were included (Figure 1). The baseline characteristics of each participant are presented in Table 1. During the follow-up period, there were 1822 individuals who survived and 658 who died. The survivor group consisted of 44.5% males and 55.5% females. 15.8% of the population identified as Mexican American, 1.9% as other Hispanic, 53.4% as non-Hispanic white, 25% as non-Hispanic black, and 3.8% as belonging to other race. In regards to educational attainment, 45.4% of the population had education beyond high school. Concerning income, 40.3% of households had an income to poverty ratio greater than 3. Of the participants in survival group, 62.1% reported consuming alcohol, 48.9% smoked cigarettes, 47.3% had hypercholesterolemia, 17.8% suffered from diabetes, 14.2% from cardiovascular disease, 5.3% from stroke, 18.6% from lung disease, 4.1% from liver disease, and 3.3% from failing kidneys. In the group that experienced death, 55.5% were male and 44.5% were female. By race, 12.3% identified as Mexican American, 1.1% as other Hispanic, 65.5% as non-Hispanic white, 19.3% as non-Hispanic black, and 1.8% as other race. In terms of educational attainment, 34.2% possessed a high school education or higher. Regarding income, 24.8% had a ratio of family income to poverty exceeding 3. Of all participants in all-cause death group, 59.7% reported alcohol consumption, 62.8% reported smoking, 52.4% had hypercholesterolemia, 28.4% had diabetes, 37.8% had cardiovascular disease, 18.1% had a history of stroke, 26.1% suffered from lung disease, 4% experienced liver disease, and 10.9% had failing kidneys.

The study examined the association between fruit consumption and all-cause mortality. The findings from splines smoothing plots analysis, as depicted in Figure 2, illustrated a decreasing trend in the risk of all-cause mortality when consuming apples and bananas three to six times per week. However, no significant changes in the risk of all-cause mortality were observed for other fruits, regardless of the intake levels.

The findings from the multivariable adjusted Model 3 reveal that individuals who consume apples or bananas three to six times per week have a reduced risk of all-cause mortality by 40% (HR = 0.60, 95% CI: 0.45–0.78, p < 0.001) and 24% (HR = 0.76, 95% CI: 0.59–0.97, p = 0.027), respectively, compared to those who consume them less than once per month (Table 2). However, no statistically significant correlation was found between other frequency of consumption of apples and bananas and the risk of all-cause mortality. Furthermore, it is worth noting that no significant association is identified between the consumption of other fruits (including pears, pineapples, and grapes) and all-cause mortality.

Subgroup analyses revealed that the association between apple consumption and all-cause mortality among individuals with hypertension remained unaffected by the presence or absence of comorbidities (all p values for the interaction were > 0.05) (refer to Table 3). Over the course of a 10-year follow-up period, our findings demonstrated that hypertensive individuals who consumed apples three to six times per week had a significantly lower risk of mortality compared to those who consumed them less than once a month. This observation is substantiated by the risk ratio results for all-cause mortality derived from the fully adjusted model, as illustrated in Figure 3 (log-rank test p < 0.05).

In a subgroup analysis, it was observed that the association between banana consumption and all-cause mortality among individuals with hypertension remained unaffected by the presence or absence of comorbidities (all p values for the interaction were > 0.05) (refer to Table 4). By employing COX regression analyses and accounting for confounding factors, the risk ratio for all-cause mortality was calculated. The results demonstrated that hypertensive individuals who consumed bananas three to six times per week had a significantly lower risk of mortality compared to those who consumed bananas less than once per month. The results are shown in Figure 4 (log-rank test p < 0.05).

Further research was conducted to investigate the relationship between the consumption of a combination of apples and bananas and all-cause mortality rates. Participants were categorized into nine groups based on their frequency of consuming apples and bananas. Weekly consumption of apples and bananas was recorded as either three to six times, greater than three to six times, or less than three to six times. The Cox proportional hazard model was used to assess the risk of all-cause mortality. The findings (Figure 5; Supplementary Table S2) suggest that consuming apples and bananas three to six times per week significantly reduces the risk of all-cause mortality compared to consuming them less than three to six times per week (HR = 0.57, 95%CI: 0.39–0.84, p = 0.005). However, other consumption frequencies did not show a significant reduction in the risk of all-cause mortality.

An important strength of this study is that it examines the association between fruit consumption and all-cause mortality within a hypertensive population, thereby providing a unique contribution to existing literature that has primarily focused on the general population. Moreover, our study boasts a substantial and diverse sample size, encompassing participants from different regions of the United States, varying ethnic backgrounds, and diverse socioeconomic statuses. The selection of the NHANHS database was based on its reputation for providing high-quality data and its recognition within the field. However, the scope of our study is constrained as it solely focused on examining the effects of apple and banana consumption on all-cause mortality, without considering the associations with specific causes of mortality such as cardiovascular disease, cancer, or chronic conditions. The researchers assessed the fruit intake of participants, which may introduce random measurement errors due to variations in individual diets over time. Additionally, observational studies may be subject to bias. In our study, we adjusted for covariates such as education level, poverty rate, and smoking to account for potential confounding factors. Although we made diligent attempts to account for all potential confounding variables and mitigate the impact of covariates in our analysis, it is plausible that unidentified or unmeasured factors could have influenced our findings.