Abstract:
Background: The 2014 WHO Noncommunicable Diseases Country Profiles [1] shows that
7% of New Zealanders die from respiratory diseases. Particularly, the Pacific Island ethnic
group is suffering deeply from these diseases. Compared to other ethnic groups, the Pacific Island
ethnic group has higher hospitalization and mortality caused by respiratory diseases. The
Pacific Islands Families (PIF) cohort study is an observational study which offers a reliable and
unique data source to investigate the causal effect of risks and resilient behaviours/protective
factors in early life on the lung function in early adulthood. The result of this study can help us
access feasible solutions for public health intervention.
Objective: This study focused on finding the impact of modifiable nutritional risk and resilience
factors in early life on the lung function in early adulthood for the Pacific Island ethnic
group.
Methods: This study was based on data collected in Pacific Islands Families (PIF) cohort,
focusing on respiratory health at age 18. We assigned variables collected during earlier assessments
to 11 domains of risk and resilience factors. To make the result more interpretable, we
applied three methods to further reduce the dimensions: combination of related variables into
a single variable, selection of variables by subject-matter experts, and combination of variables
into factor scores using factor analysis. Factor analysis was only implemented in the nutrition
domain, which is the focus of this work. Before executing factor analysis, in order to make the
food consumption comparable across measurement waves, we unified their unit to daily portions
and classified them into 12 common food categories across measurement waves. In our
study, Exploratory Factor Analysis (EFA) was employed to explore the underlying structure of
food categories and to decide how many factors (eating patterns) were needed, prior to generating
factor scores. In the following steps, we fitted a measurement invariance model, which is a
multiple group model from Confirmatory Factor Analysis (CFA). This step aimed at estimating
the loadings of food categories. This approach was selected to guarantee that the loadings were
invariant across measurement waves and could be computed uniformly across any data set.
Meanwhile, the factor scores based on these loadings should have the same invariant feature.
We used weighted sum scores to compute the factor scores in this paper. This coarse method
upholds the invariance of the factor scores and reflect the impact of food category loadings
on factors (eating patterns) in the factor scores. For causal inference, we used a causal diagram
elicited from subject-matter experts to visualize the causal paths amongst the selected
exposures, and implemented semi-parametric regression models (linear regression model and
relative risk model) to obtain the causal results of nutrition factor scores on respiratory outcomes
(FEV1 adjusted for height and sex, FEV1 Z-score, and FEV1 % predicted). It is worth
noting that the response of relative risk model was the indicator based on the cutting point
(-1.64) of healthy lung function in FEV1 Z-score. Since the PIFS cohort has suffered from
attrition, possible selection bias needed addressing. To do so, we generated weights based on
the baseline characters from the original birth cohort to reduce selection bias. In the last step,
we computed population attributable fraction (PAF) of the nutrition factor scores to estimate
the protected fraction of the healthy lung function due to nutrition at the population level, and
also showed how the PAF is subject to the change of location of nutrition factor score in the
particular eating pattern.
Results: In this study, we found that the eating patterns were basically align across all measurement
waves with some differences. Amongst them, from the result of linear models, the
”Fruit and vegetables” eating pattern at 9 years had statistically and clinically significant significant
causal effects on the healthy lung function in early adulthood. The higher nutrition factor
scores in this eating patterns, the better lung function in the early adulthood. We estimated that,
on average, one added portion per day of ”Fruit and vegetables” at 9 years will increase FEV1
Z-score by 0.25 units (95% CI: 0.00 - 0.43 units) or FEV1 % predicted by 2.94 percentage
points (95% CI: 0.00 - 4.99 percentage points) or lung volume by 120 millilitre (95% CI: 0 -
210 millilitre). Furthermore, the PAF of healthy lung function showed the causal effect of the
”Fruit and vegetables” eating pattern at 9 years was also statistically significant at the population
level. The results told that the consumption pattern of ”Fruit and vegetables” at 9 years is
accountable for 11 percentage points of the prevalence of healthy lung function (95% CI: 0 - 19
percentage points), compared to o consumption at all. It offered a feasible way to enhance the
healthy lung function prevalence amongst Pacific Island youth by a public health intervention
- increases the average daily intakes of ”Fruit and vegetables” at 9 years.
limitations: 1. Nutrition factor scores used in the study may not be completely compatible
with eating patterns from all measurement waves; 2. Some information was lost when unifying
the food categories amongst all measurement waves in the study; 3. There may be some
missing exposures in the causal diagram.
Strengths: 1. Nutrition factor scores were comparable over all measurement waves as their
unit was unified; 2. As nutrition factor scores were expressed in daily portion, the PAF obtained
in our study actually revealed how the prevalence of healthy lung function can be affected by
a change in the number of daily portions of food consumption in the particular eating pattern;
3. The causal diagram was reviewed by experts in the relevant areas, so the generated results
should be nearly unbiased; 4. Inverse probability weight (IPW) was used to guarantee a certain
degree compensate for the impact of the selection bias.
Future research: We can 1. rerun the factor analysis on the 9-years measurement wave without
following the early paper, and rebuild the models and recompute the PAF based on new
nutrition factor scores; 2. use other methods to obtain the different weights; 3. examine various
ways to change the distribution of the location of nutrition factor scores to reveal how the
PAF of healthy lung function varies over different situations.