1 Introduction

Diabetes, a chronic metabolic disorder characterized by elevated blood sugar levels, can result in various complications, such as cardiovascular disease, kidney ailments, eye disorders, and neuropathy. These complications can severely impact patients’ quality of life and longevity. As per the World Health Organization, 462 million individuals globally were living with diabetes in 2019, and this figure is projected to rise to 700 million by 2045. Diabetes not only imposes a significant burden on patients but also places considerable strain on healthcare systems and the broader economy. Consequently, effective diabetes management is a critical public health priority, necessitating various medical interventions, including drug therapy, dietary control, physical activity, blood glucose monitoring, education, and training, to lower blood glucose levels, reduce the incidence of complications, and enhance patients’ self-management capabilities and life satisfaction.

Evaluating the efficacy of medical interventions is a central aspect of medical research, as it aids in determining whether the intervention has met its intended objective, whether there are any adverse reactions or side effects, whether it is cost-effective, whether it is suitable for various populations and environments, and whether it is ethical and legal [1]. Numerous methods exist for assessing the effectiveness of medical interventions, with statistical methods being the most frequently used and crucial among them. These methods can aid in developing rational research protocols, gathering and analyzing data, drawing trustworthy conclusions, and promoting and implementing them. The selection and application of statistical methods must consider factors such as purpose, type, subject, variable, data, hypothesis, and limitations to ensure the research’s effectiveness, credibility, robustness, and adaptability. Using diabetes management as an example, this article outlines some commonly employed statistical methods, including randomized controlled trials, meta-analyses, propensity score matching, and causal inference.

2 Randomized controlled trials

2.1 Concept of the randomized controlled experiment

Randomized Controlled Trials (RCTs) are a widely employed method for assessing the effectiveness of medical interventions. RCTs represent an experimental study design that randomly assigns participants to either intervention or control groups, and then compares the difference in outcome variables between these groups before and after the intervention, thus enabling the inference of the intervention’s efficacy. The benefits of RCTs include their ability to effectively mitigate the influence of confounding factors, enhance the credibility of causal inference, and facilitate statistical analysis and conclusion drawing. However, the drawbacks of RCTs are multifaceted. For one thing, RCTs need a large pool of subjects, time, resources, and manpower. For another, there are potential issues with randomization failure, attrition, non-compliance, selection bias, etc. Furthermore, RCTs may also be subject to ethical and practical constraints, making it challenging to generalize their findings to real-world clinical and social settings.

2.2 Cases of randomized controlled trials

Numerous randomized controlled trials have been conducted on diabetes management. For instance, the UK Prospective Diabetes Study (UKPDS) is a prominent investigation that enrolled 5,102 newly diagnosed patients with Type 2 diabetes between 1977 and 1997, randomly assigning them to various drug treatment groups (insulin, sulfonylureas, high-dose metformin, etc.) and conventional treatment groups [2]. Comparisons were made in terms of blood glucose control, microvascular and macrovascular complications, and mortality. The findings of this research indicate that stringent blood glucose control can notably reduce the risk of microvascular complications; however, its influence on macrovascular complications and mortality is minimal. Moreover, different drug treatments exhibit varying effects and side effects [3]. This study offers crucial evidence and direction for diabetes drug treatment and paves the way for subsequent research.

3 Meta-analysis

3.1 Concept of meta-analysis

Meta-analysis is a statistical technique. It integrates the findings of a large number of studies, so as to enhance the statistical significance of individual investigations, address disparities among various studies, identify potential influencing factors, and finally establish more consistent and reliable conclusions. The process of conducting meta-analysis involves several steps, including defining the objective and parameters of the study, conducting a literature research, and selecting relevant studies.

3.2 Cases of meta-analysis

Numerous meta-analyses have been conducted on diabetes management, for instance, Clar et al. performed a meta-analysis in 2001 to compare the impacts of various diabetes education interventions on blood glucose control. By screening literature from 1966 to 1998, they identified 72 eligible studies involving 10,639 individuals with diabetes. Then, they categorized these diabetes patients into four types of educational interventions: namely individualized education, group education, combined individual and group education, and other educational approaches. Utilizing a random effects model, they calculated the standardized mean difference (SMD) in glycemic control for each educational intervention and conducted sensitivity and subgroup analyses. Their findings revealed that all types of educational interventions significantly enhanced glycemic control, with the combined individual and group education demonstrating the most favorable outcome (SMD=-0.51), followed by individualized education (SMD=-0.32), group education (SMD=-0.22), and other educational interventions yielding the least favorable results (SMD=-0.12). They also determined that the efficacy of educational interventions was influenced by various factors, such as study quality, duration of intervention, frequency of intervention, content of intervention, intervener, age of patient, and type of patient [4]. Their study offers valuable insights and recommendations for the design and execution of diabetes education interventions.

4 Propensity score matching

4.1 Concept of propensity analysis matching

Propensity Score Matching (PSM) is a method of matching based on the propensity score, which is the conditional probability of a subject receiving an intervention, determined by their observational characteristics. The objective of propensity score matching is to eliminate the confounding effect of observational characteristics on outcomes by matching the propensity scores of the intervention and control groups, allowing for the estimation of the Average Treatment Effect (ATE) or the Average Treatment Effect on the Treated group (ATT) [5]. The benefits of propensity score matching include reducing the dimension of matching variables, simplifying the matching process, enhancing matching quality, and avoiding issues related to over-fitting and multicollinearity. However, propensity score matching also has its drawbacks, such as the requirement to satisfy the hypothesis with strong ignorability, which posits that, given the propensity score, the allocation of the intervention is independent of the potential outcome. This is challenging to test and may also be influenced by hidden variables. Moreover, propensity score matching necessitates the selection of an appropriate propensity score model, matching method, and matching quality test, all of which can impact the effectiveness and efficiency of the matching process [6].

4.2 Cases of propensity analysis matching

Numerous propensity score matching studies exist for diabetes management. For instance, Ali et al. performed a propensity score matching investigation in 2013 to contrast the disparities in blood glucose control, medical expenses, hospitalization frequency, length of stay, and other parameters between Type 2 diabetes patients who participated in and those who abstained from the Diabetes Self-management Education (DSME) program. By screening medical records from 2008 to 2010, 1326 subjects who engaged in DSME and 1326 subjects who refrained from DSME were identified. Then, the propensity scores of each patient were estimated by employing a logistic regression model with gender, age, race, insurance type, comorbidities, and other variables as covariates. Subsequently, propensity score matching was executed using the one-to-one greedy matching method, with a caliper of 0.2 standard deviations, resulting in 1067 valid matching samples. Utilizing the T-test and Chi-square test, the researchers examined the covariate balance between the two groups before and after matching and observed a significant improvement in covariate balance post-matching, with the standardized deviation of propensity score reducing from 0.472 pre-matching to 0.039 post-matching. Through a two-sample T-test, they assessed whether significant disparities existed in outcome variables between the two groups of patients post-matching. The study revealed that subjects enrolled in the DSME program exhibited notable advantages in blood glucose control, medical costs, hospitalization frequency, hospitalization days, etc., suggesting that the DSME program positively impacts the clinical and economic outcomes of patients with Type 2 diabetes [7]. This investigation provides compelling evidence and corroborates the effectiveness and benefits of diabetes self-management education.

5 Causal inference

5.1 Concept of causal inference

Causal inference is a statistical technique aimed at identifying and estimating causal effects, which focuses not only on the correlation between two variables but also on the impact of one variable on the other. The challenge of causal inference stems from the inability to observe the results of the same subject in different intervention states simultaneously, thus precluding direct comparison of the intervention effects and necessitating the use of assumptions and methods to make inferences. There are various approaches to inferring causality, with the most common being the Potential Outcomes Model (POM), which divides the outcomes of each subject into two possible states, one with the intervention and the other with the non-intervention, and defines the causal effect as the difference in outcomes between the two states. POM-based approaches can be categorized into design-based and model-based approaches, with the former primarily relying on the design of randomized experiments and the latter primarily on the assumptions of statistical models. The advantage of the design-based approach is its ability to effectively eliminate the influence of confounding factors and enhance the credibility of causal effects, albeit with difficulties in practical implementation, and potential issues such as randomization failure, non-compliance, and interference [8]. The advantage of the model-based approach is its ability to utilize observational data for causal inference, conferring a broader scope of application. However, it necessitates the satisfaction of stringent assumptions, such as strong ignorability, commutativity, and stability. This can be difficult to verify and may be subject to interference from hidden variables or mediating variables [9].

5.2 Cases of causal inference

There are many causal inferences in diabetes management. For instance, Suissa et al. performed a causal inference analysis in 2012 to compare the risk of cardiovascular disease in patients with Type 2 diabetes who utilized and refrained from using metformin. By utilizing a medical database in Quebec, Canada, the researchers identified 122,621 patients who initiated metformin use and those who did not use it between 2000 and 2009, including 65,742 patients who took metformin and 56,879 patients who did not. They employed a Cox proportional hazards model to estimate the causal effect of metformin on cardiovascular disease, controlling for variables such as age, sex, comorbidities, and other drug use as covariates. Additionally, they considered time-dependent covariates, including cumulative metformin dose and patient survival status. Their findings indicated that patients utilizing metformin exhibited a significant reduction in the risk of cardiovascular disease, with a hazard ratio of 0.82 and a 95% confidence interval of 0.75 to 0.89. This effect was exacerbated by the cumulative dose of metformin, reaching 0.54 when the cumulative dose exceeded 40,000 mg. The 95% confidence interval was 0.40-0.73. This study provides compelling evidence and corroboration for the role of metformin in preventing cardiovascular disease in patients with Type 2 diabetes [10].

6 Conclusion

The objective of this article is to present a comprehensive overview of frequently utilized statistical methodologies and their applicability in assessing the impacts of various medical interventions. Using diabetes management as a case study, this paper delves into the intricacies of randomized controlled trials, meta-analyses, propensity score matching, and causal inference. The findings of this approach are summarized in the section dedicated to randomized controlled trials when evaluating the efficacy of diabetes management across diverse treatment regimens. In the meta-analysis section, the author compiles the results from various studies, deriving comprehensive insights into the intervention’s effects. The propensity score matching section uncovers that this method possesses the ability to minimize bias in evaluating intervention impacts and furnishes empirical evidence with tangible outcomes. In the causal inference section, the author discusses the establishment and reasoning techniques of causal relationships, arriving at causal conclusions regarding the effects of diabetes management interventions. However, a limitation of this paper is the insufficient exploration of some statistical methods, such as causal inference, where further research is warranted. Future studies could adopt a more expansive perspective to assess the impact of medical interventions, concurrently examining it in conjunction with other factors to enhance the accuracy and practicality of the research.