healthmedicinet

Studies included

We obtained 9229 records from the search performed. Of these, 32 met the inclusion
criteria (Fig. 1). Ten of these studies were missing standard error for the main meta-analysis and
therefore were only used in some subgroup analysis (Table 1). Studies included date from 1993 to 2014 and represented four locations: United
States, Canada, United Kingdom and Belgium. All studies vary in sample size and in
length of follow-up. One study reported an overall not statistically significant mean
weight loss in the sample 14] whereas almost all the other studies reported an overall statistically significant
mean weight gain. A summary of included studies can be found in Table 1 and a summary of all results, with results of heterogeneity tests, can be found in
Table 2.

Fig. 1. Sampling and selection of articles on weight gain in university students

Table 2. Summary of random-effects meta-analysis estimates by type of analysis

Mean weight change

To perform a meta-analysis, the mean weight change as well as the standard error or
standard deviation must be reported. Only 14 studies reported the necessary data.
After contacting authors, we had sufficient data for a total of 22 studies (Table 1: Analysis 1a), accounting for a sample of 5549 students. In this analysis we observed
a statistically significant weight gain of 1.36 kg (CI: 1.15–1.57, I²?=?85.1 %) over 6 weeks to eight months (Fig. 2). We also performed a meta-analysis weighted on sample size which yielded a mean
weight gain of 1.21 kg (CI: 1.12–1.30, I²?=?85.2 %).

Fig. 2. Meta-Analysis of mean (95 % CI) weight change (kg) from baseline to follow-up in first
year university students; studies reporting standard errors. The overall mean weight
change is 1.36 kg (CI: 1.15–1.57)

Three studies had reported enough data to allow us to impute a standard error. When
these were included in the meta-analysis (Table 1, Analysis 1b), along with with 22 studies which reported standard error data, the
mean weight gain was 1.35 kg (CI: 1.15–1.55). This pooled mean was not significantly
different. Seven studies could not be included in the meta-analysis of mean weight
change as they did not report a standard error and we were unable to obtain the missing
data from authors. We conducted a Wilcoxon-Mann-Whitney test to compare the reported
weight change in studies included and not included in the meta-analysis. These were
not significantly different (P??0.61). To avoid using imputed data and to deal with missing data, the subsequent
subgroup analysis only used studies which reported standard errors (Table 1, Analysis 1a).

Heterogeneity and bias analysis

In all meta-analyses, heterogeneity was high (above 80 %). We investigated heterogeneity
through a Galbraith plot; studies were very different but no single study contributed
significantly more to heterogeneity than others. The large differences in sample size
and in differences in length of follow-up did contribute to heterogeneity. It is to
note, that the studies had different gender composition and were conducted in different
countries, which likely adds to the heterogeneity. We further performed Wilcoxon-Mann-Whitney
tests to compare studies which were included and not included in the meta-analyses
and sub-group analyses. In every instance, there was no statistical difference. We
also analysed the potential for publication bias through a funnel plot and found a
symmetric plot indicating likely low publication bias.

Subgroup analysis by location

We further investigated weight changes according to location of study; 15 studies
were conducted in the United States, five in Canada and one each in the United Kingdom
and Belgium. The United Kingdom and Belgium had lower weight gain than in the United
States and Canada; but inference cannot be made due to the small sample of studies.
Canadian and US studies did not significantly differ in weight gain with Canada having
a pooled mean of 1.71 kg (CI: 1.04–2.38, I²?=?86.5 %) and the United States, 1.32 kg (CI: 1.08–1.56, I²?=?84.6 %).

Subgroup analysis by measurement method

In this analysis (Table 1, Analysis 1a), 18 studies measured weight change objectively while four relied on
self-report. Running a sub analysis, studies using measured weight 1.45 kg (CI: 1.21–1.69,
I²?=?80.9 %) were not significantly different than the pooled mean of those using self-reported
weight, 1.04 kg (CI: 0.63–1.44, I²?=?88.6 %). We excluded two studies who used different weight measuring methods at
baseline (self-report) and follow-up (measured) 26], 27].

Subgroup analysis by study retention rate

We investigated the effects of the percentage of students, from the initial sample,
who completed the study. The weighted retention rate was 57 %, although half of the
studies had a retention rate higher than 80 %. The mean weight gain was not significant
different between studies having a low retention rate (less than 40 %), medium retention
rate (40–80 %) and high retention rate (above 80 %).

Subgroup analysis by study length

The average length of the studies was five months, with a range from 6 weeks to 8 months.
We investigated whether the length of studies had an impact on the reported weight
change. We stratified by length of study: four months or less (representing one university
term) and more than four months. Studies which had two data collection point during
the year could be included a maximum of once per strata. The results showed that studies
of longer length reported higher weight gain, 1.47 kg (CI: 1.13–1.81, I²?=?89.6 %) than shorter studies, 1.24 kg (CI: 0.96–1.53, I²?=?77.3 %).

As it is often reported that weight gain in 1st year university students predominately
happens during the first term, we also assessed whether weight gain occurs predominately
during the first term or whether it is constant over the year. Six studies had two
different time points of follow-up, with reported SD. In these studies over the first
four months, students gained an average of 1.24 kg (CI: 0.87–1.61, I²?=?81.6 %) and by the end of the academic year, they had gained on average 1.76 kg
(CI: 1.32–2.21, I²?=?82.6 %). When we conducted a meta-regression, the length of follow-up was significant
(p??0.05) to predict higher weight change; the univariate R² was 27.1 %.

Weight gain in weight gainers (Analysis 2)

The overall mean weight change is affected by outliers and students losing weight.
We therefore examined the percentage of students gaining weight through a weighted
average of 16 studies (Table 1, Analysis 2a). A majority of students, 60.9 %, gained weight during freshman year,
although studies did not have the same definition of weight gain, which is a source
of heterogeneity. Some required a minimum threshold weight gain while others had none.
To further investigate weight gain, we calculated weight gain in those who did gain
weight only in order to get a sense of the magnitude of weight gain this subgroup.
Nine studies reported the average weight gain in weight gainers but only five reported
the standard deviation, required for inclusion in a meta-analysis (Table 1, Analysis 2b). The subpopulation of weight gainers gained 3.38 kg (CI: 2.84–3.92,
I²?=?89.5 %), significantly higher than the overall pooled mean weight gain in the general
first year student population. Three studies (Table 1, Analysis 2c) also reported the percentage of students gaining the “Freshman-15”
15 lb (6.8 kg). The weighted average of these studies showed that 9.3 % of 1st year
students gained at least 6.8 kg.

Analysis by gender

From the 32 studies included in this review, ten studies were conducted only with
females while one study was solely conducted with males (Table 1). Twelve other studies reported weight change stratified by gender. On average, studies
which had a mix sample had a higher percentage of female participants. Due to lack
of reporting of standard deviations, only 14 studies could be used for a meta-analysis
by gender (Table 1, Analysis 3). Females and males did not differ in their weight gain with each gaining
respectively on average 1.34 kg (CI: 1.02–1.65, I²?=?79.9 %), and 1.43 kg (CI: 0.90–1.97, I²?=?83.5 %).

Analyses by quality of study

In general, the quality was adequate with most studies ranking 5 points out of 7.
Ten studies were of high quality (6–7 points), 21 of medium quality (4–5 points) and
one of low quality (3 or less points). Studies of medium or low quality did not have
a significantly different mean weight change than studies of high quality (p??0.71). In general, studies had poor recruitment strategies and did not have a representative
university sample.