4. Discussion
We developed a clinical risk prediction score to predict the possibility of serious GI complications in elderly patients receiving NSAIDs using geriatric population data from South Korea. By demonstrating the good performance of the prediction model through external validation, we confirmed that it is not limited to patients with arthritis and can be applied to the general elderly population that uses NSAIDs. Because the factors contributing to the risk of GI bleeding are different in the elderly and young populations using NSAIDs, risk prediction models specific to the elderly population should be considered. To the best of our knowledge, this is the first study to predict GI complications associated with NSAID use in the elderly population.
In our study, high-dose NSAID use was the greatest risk factor for serious GI complications as well as a history of complicated GI ulcers. The history of complicated GI ulcers is, as already known, a potent risk factor for NSAID-induced ulcers and a critical criterion in GI ulcer prevention algorithms [17] [18]. We found that NSAID use in excess of the daily recommended dose (e.g., ibuprofen 1200 mg, naproxen 500 mg, celecoxib 400 mg) or multiple NSAID use was a risk factor similar to a history of complicated GI ulcers in the elderly population. Multiple NSAIDs may be prescribed simultaneously, which may result in patients taking high-dose NSAIDs. Multiple prescribers may prescribe different NSAIDs. Patients may use multiple pharmacies or pharmacists may not detect multiple NSAID prescriptions. Therefore, a systematic medication management system for the elderly is required.
Our study identified new risk factors not included in the current guidelines [19]: male sex, very old age, and concomitant use of SSRI. Previous risk prediction studies [1] [6] for upper GI bleeding related to NSAID use have reported risk factors similar to those identified in our study, although their study populations were not specific to the elderly. In one study, age, male sex, anemia, aspirin, and anticoagulants were identified as predictors, while GI history and co-treatment with corticosteroids and SSRIs were not [6]. In another study, age; history of GI bleeding or perforation; concomitant use of corticosteroids, SSRIs, and antithrombotic agents; and male sex were identified [1]. Although these factors were included in our analysis, they were not identified as predictors. In our study, concomitant PPI or H2RA treatment with NSAID reduced the risk of serious GI complications, whereas the use of rebamipide or Artemisia herb soft extract did not. According to the guidelines, PPIs are recommended to prevent NSAID-related GI damage [17], but in practice, alternative GPAs are sometimes prescribed because of concerns about the long-term safety of PPI [12]. According to our previous study that reported the pattern of GPA prescription in the elderly (≥ 65 years) using NSAIDs, PPI, H2RA, rebamipide, and Artemisia herb soft extract accounted for 11.4%, 24.8%, 8.0%, and 6.8% of the total GPA prescriptions in Korea, respectively [20]. In a previous study investigating whether GPAs effectively prevent NSAID-related GI injury in patients with arthritis, mucoprotective agents such as rebamipide and misoprostol were effective in reducing the risk of GI bleeding in NSAID users as acid suppressants (PPI or H2RA) [12]. However, the population in this study who used NSAIDs were over 20 years of age, a population with a relatively low GI risk compared to the population in our study. In our evaluation of GPAs, only PPI and H2RA, which acted as offset factors that reduced the risk of serious GI complications, were included in the prediction model.
We developed a risk model for capturing expanded cases compared to the risk classification system of the current guidelines, which counts the number of risk factors [19]. All of the ten cases with high frequency predicted as high-risk patients had the following factors: age ≥75 years, male sex, concurrent use of antithrombotic agents, and concurrent use of corticosteroids. If avoidable factors, such as high-dose NSAIDs, concomitant glucocorticoids, and antiplatelets, were eliminated, all ten high-risk cases were no longer high-risk.
Our study evaluated the use of GPAs as risk prediction factors for NSAID users. PPI or H2RA use was not sufficient to offset serious GI complications in all cases. According to our prediction model, men over 75 years of age using nsNSAIDs with concomitant single antiplatelet and glucocorticoid therapy were still at high risk even though they had used PPIs. Therefore, it is important to closely monitor and eliminate avoidable risk factors, if possible, when a patient is predicted to be at high risk, even when using a PPI.
The development of a convenient risk prediction score for the safe use of commonly used medications such as NSAIDs could benefit public health. The risk prediction calculator for serious GI complications developed in our study demonstrated acceptable performance and discrimination in the external validation, even without individual laboratory tests or specific information known only to experts. It is beneficial to quickly calculate the risk scores using only the patient’s prescription and severe comorbidities known to themselves. By deriving cutoff values, high-risk patients can be easily identified.
Our study had several limitations that should be considered. First, as our assessment of predictors of GI complications depends on the factors available in the claims database, there might have been unmeasured risk factors. We could not include Helicobacter pylori infection, genetic susceptibility, or social factors such as smoking and alcohol consumption. However, smoking is not a significant risk factor for GI ulcers [1], and NSAID-induced GI complications are unrelated to H. pylori infection [2]. Second, the identification of serious GI complications and risk factors from claims data might be inaccurate, as it was based on the ICD-10 diagnostic codes [21]. Third, since age information over 80 years was not provided in the external validation data, we were unable to evaluate how well our model discriminated this age group with high confidence. Fourth, our model may not be generalizable to populations other than those used to derive and validate the model [21]. Additional external validation should be conducted to generalize the results to other populations (e.g., Western countries). Fifth, some of the medications sold over the counter, including NSAIDs and H2RA, could not be identified in the claims data.