“Gestational diabetes mellitus” (GDM) is diabetes that develops during pregnancy. It can cause many health problems for pregnant women and their babies and after birth. Having diabetes during pregnancy can mean both women and their babies are at an increased risk of serious complications in future. However, the risk of GDM can be modifiable before pregnancy with several healthy habits. With this background, the researchers followed 14,437 female nurses aged 24 to 44 who enrolled in a study ‘the Nurses’ Health Study II’ to observe the association between healthy lifestyle and diabetes during pregnancy from 1989 to 2001. The researcher defined a healthy lifestyle by four factors: 1) no smoking; 2) eating healthy food; 3) exercising moderately to vigorous at least 150 minutes a week, 4) Body Mass Index (BMI)<25 (e.g. BMI is an indicator calculated by height and weight. A BMI between 18.5 and 24.9 is considered healthy weight). The researchers gained data through regular questionnaires asking the participant’s lifestyle and body weight. If anyone is diagnosed with GDM, they should report it to the researchers. During the 10-years of follow-up, a total of 823 pregnancies were related to GDM among 14,437 women. If participants adhering low-risk lifestyle are classified in “the low-risk group”. Compared with women did not have a healthy lifestyle before pregnancy, the researchers estimated how much four lifestyle factors reduce the risk of GDM. As a result, keeping healthy body weight would be substantially beneficial to reduce the risk of GDM, but other healthy lifestyle factors (that is, healthy diet, regular exercise, not smoking) are also related to a lower risk of the disease regardless of their obesity. In conclusion, a healthy lifestyle before pregnancy could be helpful to prevent diabetes during pregnancy.
a) To alleviate the effect of chance at the design stage, choosing an appropriate type of study is necessary, which depends on the questions the study is addressing. Plus, increasing the sample size of the study is a good way to reduce the chance. Lastly, it also important to choose stringent methods (i.e. questionnaire, medical records and so on) to detect outcomes.
Table 1. Information needed for calculating sample size (Bonita et al., 2006. p.53)
|Information needed for calculating sample size|
Table 2. Types and applications of epidemiologic study (Bonita et al., 2006, p.49)
|Long latent periods||++++||+++
(If historical cohort)
|Multiple effects of causes||–||++++||++||+|
|Measurements of time relationship||+||++++||–||+|
|Multiple exposures and determinants||++++||+++||++||++|
|Direct measurement of incidence||–||++++||–||–|
* – : not useful; +/++/+++: some useful; ++++: very useful
b) At the stage of analysis, utilising appropriate statistical analysis methods in response to the type of studies, type of data, the number of variables (e.g. exposure, risks), potential compounding can minimize the chance. Generally, Confidence Intervals (CI) and P-value can show the precision and significance of results, respectively. CI provides information on the magnitude of effect and indicates the precision of estimated value through interval width. If the interval is narrow, the true population parameter reasonably lies within the range with a certain confidence. P-values are calculated from the null hypothesis (H0) to assess associations between risk factors and outcomes. if the p-value is lower than 0.05 (conventional threshold, p<0.05), there is a statistically significant association between two groups with the rejection of H0. Type 1.2 (table)
If you need assistance with writing your essay, our professional essay writing service is here to help!
c) At the design stage, the authors used a large sample size (N=116,671), which could reduce sampling errors. To alleviate measurement errors, they used the well-proven questionnaire and targeted only nurses that are familiar with health-related terminology. However, there might be over-reporting, which may lead to type 2 errors. The researchers estimated and showed a high correlation between the self-reported weight and measured weight of a random subset of cohort participants in the Boston area.
At the analysis stage, the authors estimated each factor’s RR and PARP, which is appropriate for cohort studies. Regarding variables, the authors did “multivariate model” to calculate RR and CI. The results of questionnaires are assessed by correlation analysis (Table. 2) Measured PARP were also validated by proper calculation. They reported CI for around all estimates, but not any p-values. Thus, it is not easy to assess the statistical significance of the association between protective factors and GDM.
Table 2. Examples of correlation analysis
|Physical activity as reported in one week recalls &
Physical activity as reported on the questionnaires
|Moderate to vigorous activity reported in diaries &
Moderate to vigorous activity reported on the questionnaires.
|Self-reported weight &
Weight measured by a technician among a random subset of participants in the Boston area cohort
d) The null hypothesis would be as follows: there is no difference in GDM incidence for women having a BMI greater than or equal to 35 and women who do not (Table 1). Women having BMI⪰ 35 seems have quadruple (RR=4.45) the risk of GDM compared to women who do not, with 95% confidence (95% CI 3.19 to 6.20). When the P-value is not presented but its CI excludes 1, the P value associated for the hypothesis test would be assumedly less than 0.05. Therefore, the null hypothesis will be rejected; there is a significant evidence of the association between BMI⪰ 35 and a risk of GDM.
a) Selection bias occurs when the characteristics of selected people are systematically different from those who are not (Henderson and Page, 2007). Therefore, the result look significantly meaningful even though there is no sufficient evidence of the association between the exposed group and the non-exposed group.
Our academic experts are ready and waiting to assist with any writing project you may have. From simple essay plans, through to full dissertations, you can guarantee we have a service perfectly matched to your needs.
Information bias can occur when methods are not appropriate to collect information from the participants, thereby some information about exposure or disease can be distorted (Althubaiti, 2016). There are several information biases such as recall bias, surveillance bias, reporting bias, misclassification bias, interview bias and so on.
Reverse causation occurs when an exposure alters an effect of the outcome condition not a risk factor, i.e. participants change their lifestyle, diet after a disease developed or after obtaining information on the disease. When it occurs, the relationship between cause and effect cannot be significant. (Shakiba et al., 2016)
b) Firstly, since female nurses were included, most participants were aware of various chronic diseases including GDM. Thus, they might take care of their health more than the general population. For example, 27% of people that were overweight or obese were included despite 60% prevalence of being overweight or obese among US women. In this case, the PARP of parous nurses might underestimate the burden of unhealthy behaviour on the risk of GDM in the general population. Next, the imbalance between races (93% white) can hinder to represent the true population. If GDM or BMI is related to genetics, the RR and PAR of the GDM can be changed.
c) Measurement bias can affect exposures of the study. Since the study used self-report (physical activity, BMI, diet, weight, smoking) as measurements, the participants are subject to remember information not clearly; the accuracy of memory would vary between individuals. Some methods for data collection were not precise or improper: for scoring dietary quality, scoring the level of intake can be ambiguous due to strong subjectivity. If this bias occurred severely in both groups, non-differential misclassification bias can occur and the RR of factors can be attenuated and go null (closed to 1). As a result, evidences showing a significant association between healthy lifestyle and GDM might be attenuated. To reduce this bias, the researcher used well-validated questionnaires as “golden standards”. For example, for overall dietary quality, they utilised the Alternate Healthy Eating Index-2010 based on the US Department of Agriculture Healthy Eating Index. They also excluded not-returned questionnaires for food frequency, unrealistic reports, questionnaires left more than 70% blank. Furthermore, for physical activity, they estimated the correlation between what was reported in once a week recalls and what was reported on the questionnaire. For self-reported weight, the authors measured a random subset of the study cohort and compared both results.
d) The participants, especially healthcare professionals, can reevaluate themselves and change their lifestyle or diet during the research. When people eating poorly were told they have at high risk of GDM, for example, many might change their diet. Such a changed habit could lead to the result look as if a healthy diet is more likely to contribute to reducing the risk of GDM. Thereby, it can underestimate the incidence of GDM compare to the true population.
To address this bias, the researcher excluded pregnancies after GDM since women having experienced GDM likely changed their diet or lifestyle for the next pregnancy. Additionally, the researcher conducted large prospective cohort for a long term of follow-up (Lawlor et al., 2006) to reduce the potential effect where obese people may prefer to eat low fat or low sugar food.
a) Confounding is a 3rd factor affecting both exposures independently to the exposure as risk factors and the occurrence of disease as outcomes (Bonita et al., 2006. P.55) Confounder should be associated with exposure, but not a result of it. At the same time, confounder should be a risk factor of the disease.
b) At the design stage, there are three methods to control confounders: Randomization, restriction and matching. Randomisation evenly distributes potential confounders between all groups or individuals. Restriction is a method to exclude potential confounders. Matching is a method to match groups with potential confounders such as age, sex and socioeconomic status.
c) At the analysis stage, stratification, standardisation and multivariable analysis can be used to assess the effect of confounding. Stratification is a way of arranging people into different layers based on potential confounders. Standardisation is a method for managing different underlying characteristics of the groups, such as gender, age. Multivariate testing, as a statistical modelling method, can be used for estimating the impact of confounding variables while controlling other variables simultaneously.
d) To undermine the effect of confounding, the authors used ‘restriction’ at the design stage of the study and ’multivariate analysis’, ’stratification’ during the analysis of the study. The researcher restricted the analysis to women with potential confoundings such as a history of GDM, type 2 diabetes, cancer, cardiovascular disease events.
At the analysis stage, they stratified the low-risk group with risk factors (family history of diabetes, parity, race/ethnicity, age, overweight) and showed coherent results (i.e. a significant association between a low risk life style and a lower risk of GDM) regardless of these risk factors. Also, they established the multivariable models to estimate associations between four major factors and the risk of GDM with the relative risk and 95% CI. Plus, the data were adjusted for age, parity, family history of diabetes, history of infertility, race/ethnicity, questionnaire period, total energy intake, and alcohol intake.
e) Potential confounders would be other diseases not mentioned in the study. A disease related to a level of Thyroid Stimulating Hormone (TSH) can act as a confounder (Fig.1). There is a significant association between TSH and BMI, but TSH is not on the causal pathway between BMI and the disease. Also, a low level of TSH is known to be a risk factor of the disease. When If women with TSH disorders adhered to a healthy lifestyle, the incidence of GDM in a low group would be higher than women without. The authors did not exclude this potential confounder. Therefore, to get more accurate and reliable result, the authors should control TSH disorders or other diseases.
Fig 1. Potential confounding – Levels of Thyroid Stimulating Hormone
5. DESIGN A CASE-CONTROL STUDY
a) All participants are women attending antenatal services in 14 hospitals in each state. Cases are recruited women diagnosed with GDM for the first time during the period of the study. Cases are selected randomly utilising a systematic random sampling program. Controls are selected women attending the same hospital with cases during the same period. Controls are without a diagnosis of GDM.
One strength of the design is that it includes various socioeconomic classes and races to reduce confounding. Also, due to selecting controls from the same hospitals with the cases, it can save time and money. One of the weaknesses is that selecting the controls representative of the population without the disease would not be easy. Besides, selecting controls identical with cases in all parts except the disease would be a challenge.
b) Both questionnaire and medical records collect data. The questionnaire consists of items asking about physical exercise, healthy diet, smoking and medical records is used for BMI records and the diagnosis of the disease. The advantage of this method is that it is easy to gain the information faster with a relatively low budget. Also using medical records can improve the accuracy of the data.
However, the information bias, especially recall bias, would occur frequently due to the use of questionnaires. Also, it is not easy for the participants to recall all information regarding exposure, but thereby misclassification non-differential bias can occur since exposure status can be incorrect in both groups.
c) Compared to the cohort study, the case-control study can assess the association between various risk factors and outcomes simultaneously. Besides, conducting a study would be faster and save more money than a cohort study. Therefore, it is useful to understand the association between exposure and diseases in early stage of the study.
d) First, identifying the effects of time on the disease can be difficult due to a relatively short study period. There is also a risk of over-matching which can mask findings. Furthermore, calculating the relative risk and incidence or prevalence of disease can be difficult. Also, compared to cohort study, recall bias and selection bias can more occur.
6. A RECENT STUDY ON THE TOPIC
A study by B.E. Feleke et al (2018) is an unmatched case-control study to identify determinants of GDM in Ethiopia (Feleke, 2018). This study published in 2018 in The Journal of Maternal-Fetal & Neonatal Medicine. This study was conducted on all pregnant women attending antenatal care attendants from five referral hospitals of Amhara regions from January 2016 to June 2016. Case (n=567) and controls (n=1690) were recruited starting from 24 weeks of gestation and 32 weeks of gestation, respectively. Women with GDM were excluded from the study. The data were collected using interview technique, measuring anthropometric indicators and taking the blood sample to measure their blood sugar level. The results indicated that the risk of GDM was increased by family history of DM, history of gynaecological disease or surgery, baseline disease, not a healthy lifestyle and literacy. However, physical activity significantly reduced the risk of GDM. Therefore, this study supports the present study with the significant role of exercise to reduce the risk of GDM. Also, this case-control study assessed various exposures, and some exposures can be considered for the design of the cohort study to reduce confounding: e.g. gynaecological disease or surgery and literacy.
Table 3. Determinants of gestational diabetes (Feleke, 2018)
|Risk Factors||AOR* [95%CI]|
|history of abortion||5.05 [2.65–9.63]|
|family history of diabetes mellitus||8.63 [5.19–14.35]|
|chronic hypertension||4.63 [1.27–16.86]|
|dietary diversification score||2.96 [2–4.46]|
|regular physical exercise||0.03 [0.01–0.04]|
|history of infertility||6.19 [1.86–20.16]|
|history of Caesarean section||3.24 [1.58–6.63]|
|previous history of GDM||8.21 [3.18–21.24]|
|previous history of intrauterine fetal death||3.96 [1.56–10.04]|
|body mass index||2.96 [2.08–4.2]|
*Attributable Odds Ratio
7. EXTERNAL VALIDITY
The study is not easily generalised since all participants were nurses in a similar educational situation. Therefore, it cannot apply to population in various socioeconomic statuses or academic levels. It is difficult to apply to various races because of the small sample size and wide confidential interval width (0.05 to 0.74).
8. RELEVANCE FOR MAKING POLICY
This study is moderate relevant for making policy due to weak generalisation and not elaborate control of bias in terms of self-reporting. Nevertheless, a program for managing BMI would be helpful to prevent GDM since when BMI, added to the combination of factors, the attributable rate affecting population increased by around 48%. Moreover, providing integrated lifestyle counselling into preconception care for women can be implemented to improve their pregnancy and birth outcomes.