Background

An adherence model is required to optimise medication management among chronic kidney disease (CKD) patients, as current assessment methods overestimate the true adherence of CKD patients with complex regimens. An approach to assess adherence to individual medications is required to assist pharmacists in addressing non-adherence.

Objective

To develop an adherence prediction model for CKD patients.

Methods

This multi-centre, cross-sectional study was conducted in 10 tertiary hospitals in Malaysia using simple random sampling of CKD patients with ≥1 medication (sample size = 1012). A questionnaire-based collection of patient characteristics, adherence (defined as ≥80% consumption of each medication for the past one month), and knowledge of each medication (dose, frequency, indication, and administration) was performed. Continuous data were converted to categorical data, based on the median values, and then stratified and analysed. An adherence prediction model was developed through multiple logistic regression in the development group (n = 677) and validated on the remaining one-third of the sample (n = 335). Beta-coefficient values were then used to determine adherence scores (ranging from 0 to 7) based on the predictors identified, with lower scores indicating poorer medication adherence.

Results

Most of the 1012 patients had poor medication adherence (n = 715, 70.6%) and half had good medication knowledge (n = 506, 50%). Multiple logistic regression analysis determined 4 significant predictors of adherence: ≤7 medications (constructed score = 2, p < 0.001), ≤3 co-morbidities (constructed score = 1, p = 0.015), absence of complementary/alternative medicine use (constructed score = 1, p = 0.003), and knowledge score ≥80% (constructed score = 3, p < 0.001). A higher total constructed score from the prediction model indicated a higher likelihood of adherence (odds ratio [OR]: 2.41; 95% confidence interval [CI]: 2.112–2.744; p < 0.001). The area under the receiver operating characteristic (ROC) curve of the developed model (n = 677) had good accuracy (ROC: 0.867, 95% CI: 0.840–0.896; p < 0.001). The validated model (n = 335) also had good accuracy (ROC: 0.812, 95% CI: 0.765–0.859; p < 0.001). There was no significant difference between the development and validation groups (p = 0.11, Z-value:1.62, standard error: 0.034).

Conclusion

The score constructed from the medication adherence prediction model for CKD patients had good accuracy and could be useful for identifying patients with a higher risk of non-adherence, to ensure optimised adherence management.

中文翻译：

---

持续用药监测：预测慢性肾病患者用药依从性的临床模型

背景

需要一个依从性模型来优化慢性肾病 (CKD) 患者的药物管理，因为当前的评估方法高估了 CKD 患者对复杂方案的真实依从性。需要一种评估个体药物依从性的方法来帮助药剂师解决不依从性问题。

客观的

为 CKD 患者开发依从性预测模型。

方法

这项多中心、横断面研究是在马来西亚 10 家三级医院进行的，对 CKD 患者使用≥1 种药物进行简单随机抽样（样本量 = 1012）。进行了基于问卷调查的患者特征、依从性（定义为过去一个月内每种药物的消费≥80%）和每种药物的知识（剂量、频率、适应症和给药）。根据中值将连续数据转换为分类数据，然后进行分层和分析。在开发组 (n = 677) 中通过多元逻辑回归开发了依从性预测模型，并在剩余的三分之一样本 (n = 335) 上进行了验证。然后使用 Beta 系数值根据确定的预测因子确定依从性分数（范围从 0 到 7），

结果

1012 名患者中的大多数患者的药物依从性较差（n = 715, 70.6%），一半具有良好的药物知识（n = 506, 50%）。多元逻辑回归分析确定了 4 个依从性的重要预测因素：≤7 种药物（构建评分 = 2，p < 0.001），≤3 种合并症（构建评分 = 1，p = 0.015），不使用补充/替代药物（构建得分 = 1，p = 0.003），知识得分≥80%（构建得分 = 3，p < 0.001）。预测模型的总构建分数越高表明依从性越高（优势比 [OR]：2.41；95% 置信区间 [CI]：2.112–2.744；p < 0.001）。开发模型的受试者工作特征 (ROC) 曲线下面积 (n = 677) 具有良好的准确性 (ROC: 0.867, 95% CI: 0.840–0.896; p < 0.001)。经验证的模型（n = 335）也具有良好的准确性（ROC：0.812，95% CI：0.765–0.859；p < 0.001）。开发组和验证组之间没有显着差异（p = 0.11，Z 值：1.62，标准误差：0.034）。

结论

根据 CKD 患者用药依从性预测模型构建的评分具有良好的准确性，可用于识别不依从性较高的患者，以确保优化依从性管理。