Predicting public (mis)use of antibiotics

How might we design effective interventions to promote the correct antibiotics use among the public?

Antibiotic resistance is a global health threat. In high-income countries, reducing unnecessary use of antibiotics is one immediate action to decelerate the speed of this resistance. While research has been conducted on knowledge of and attitudes towards antibiotics use, little is known about the overarching institutional impact on individuals’ acquisition and intake of antibiotics.

Objectives

To identify the hidden risk groups of antibiotics users among the general public
To examine how the membership is determined by intrapersonal, interpersonal, and community-level factors.

research questions

[RQ1]: Are there inductively and analytically distinct subgroups of participants based on patterns of their acquisition and intake of antibiotics?

[RQ2]: Is the class membership significantly different between countries? If yes, how much can be explained by level 1 factors and how much by level 2 factors?

[RQ3]: What systemic factors predict class membership?

[RQ4]: What individual factors and systemic factors together predict class membership?

methods

Combining four recent antibiotics modules of Eurobarometer, a latent class analysis (LCA) was conducted based on the acquisition and usage pattern of antibiotics by the representative sample from the 27 EU member states. Multilevel multinomial logistic regression was then conducted to identify individual and structural factors that predict the risk group membership.

Figure 1: Ecological perspectives based on the Social Ecological Model

Source: Rimer and Glanz (2005, p.11) and McLeroy et al., (1988, p355)

Social Ecological Model (SEM)

In the SEM, Individuals' health behaviours can be influenced by multiple factors nested in different social layers. This study uses a simplified SEM model with three layers:

Intrapersonal factors
Interpersonal processes
Community factors

analytical strategy

Exploratory and evaluative | Latent Class Analysis | Multilevel Logistic Regression | Sample size: 30K

Data preparation

To obtain statistically significant results for LCA, this study merged four datasets of recent Eurobarometer surveys.(N=109,884). After extracting individuals who have recently taken antibiotics, the final dataset yielded a total sample of 37,956 from 27 European countries.

Latent Class Analysis (LCA)

Next, I conducted LCA using a software programme R with the poLCA package to yield the unobserved risk groups among the population who have taken antibiotics by analysing not just the reason for intake but also the means of acquisition simultaneously.

Multilevel Modelling

Then, a multilevel multinomial logistic regression was run with no predictors. This is to assess whether there are significant variabilities between countries that need to be further investigated. The intraclass correlations (ICC) indicated a statistically significant difference between countries.

Logistic Regression

Further, community/policy level variables are added to the previous model as fixed effects. The next model was run by adding intra-personal and interpersonal level variables to the model as fixed effects. Finally, a multinomial logistic regression by including countries as variables to see individual effects controlled by country.

results

One low-risk group and three risk-groups emerged from the LCA (See Figure 2 below).

The properly-prescribed (low-risk) - Those with prescribed antibiotics for appropriate purposes (68% of the total sample, N=24,702)
The susceptible - In-patients who are administered antibiotics for a range of cause. (17%, N=6,273)
The mis/over-prescribed - Those with antibiotics prescribed for inappropriate illnesses/symptoms (8%, N=3,089)
The unprescribed - Those who obtained antibiotics without prescription for a range of symptoms (6%, N=2,364)

Age, financial status, and knowledge about antibiotics were found to be the important predictors of the risk-group membership.
The study also found that a nation’s level of trust in healthcare professionals, as well as income inequality, can have a significant effect on the membership of risk groups.

The properly-prescribed

68%

The

susceptible

17%

The mis/over-prescribed

The

unprescribed

Note: *(asterisk) denotes illnesses/symptoms that require treatment with antibiotics (CDC, 2019a, 2019b)Source: doi:10.4232/1.11642; doi:10.4232/1.12730; doi: 10.4232/1.13375; doi:10.4232/1.13211.

Figure 2: LCA posterior probabilities of reasons for antibiotics oral intake by latent class

conclusion

While elevating antibiotics knowledge of the general public appears to be effective, addressing the issue of trust and income inequality can contribute to the appropriate antibiotic use of individuals.

design recommendation

The results of this study indicate recommendations for designers of health intervention:

Individuals are still likely to benefit from the didactic antibiotic resistance promotions.
Individuals in risk groups are likely to trust information from family and friends more than that of doctors and nurses, designers of health interventions could leverage word-of-mouth communication beyond conventional channels in addition to a top-down approach.
It is necessary for governments to address the public trust towards healthcare professionals as well as economic inequality.
Interventions from the higher layers of the environment can play a critical role in not just slowing down antibiotic resistance but also addressing the issue of health equity as well as distrust in public health in general.

RefLection & Nest steps

While the study has examined heterogeneity among the population who have taken antibiotics among the total sample, why individuals had to take antibiotics in the first place is another important issue for the understanding of the real-world problem of antibiotic resistance. This could be followed up by qualitative studies, such as in-depth interviews, observations or diary studies.
Even though the main datasets used for the study offer a significant amount of antibiotic-related variables, some potentially important questions such as ethnicity and religiosity, have not been asked.

hypotheses tested

Unobserved risk groups

Based on patterns of response to variables regarding the acquisition of antibiotics and reasons for intake will produce more than two latent groups beyond the conventional dichotomy of rational versus irrational use of antibiotics by the general public. [Accepted]

Between-country differences of the unobserved risk group membership

There is no statistically significant variability in the latent group membership between countries. [Rejected]

Structural factors that predict the unobserved risk group membership

There is no statistically significant negative association between health expenditure and membership of the latent risk groups. [Accepted]
There is no statistically significant positive association between income inequality and membership of latent risk groups. [Rejected - there is a significant positive association]
There is no statistically significant negative association between the number of doctors available and membership of latent risk groups. [Accepted]
There is no statistically significant negative association between trust in medical professionals and the membership of latent risk groups. [Rejected[]

Individual-level factors that predict the unobserved risk group membership

Knowledge of effectiveness and risk of antibiotics have no statistically significant negative effects on membership of latent risk groups. [Rejected]
There is no statistically significant negative association between receiving antibiotics warning messages and the membership of latent risk groups. [Accepted]

Page updated

Google Sites

Report abuse