A cross-sectional study was conducted at Hospital Canselor Tuanku Muhriz (HCTM), a university hospital in Malaysia. It was formerly known as Universiti Kebangsaan Malaysia Medical Centre. It serves as a referral center for patients from Klang Valley and is also the first hospital in Malaysia that implemented a casemix system. HCTM requires coders to conduct active and ongoing diagnosis and procedure coding exercises.

This was a cross-sectional comparative study using secondary data obtained from inpatient discharge lists between January 2023 and December 2023. The aim was to assess the accuracy of clinical coding for the principal diagnoses. All patient information for the episode of admission was stored in the EMR system in the hospital database. In addition, the casemix unit at the International Centre for Casemix and Clinical Coding (ITCC) HCTM performed the diagnosis and procedure codings by extracting data from the EMR system. The database was accessed from June 1, 2024, to July 31, 2024. The authors had no access to information that could identify individual patients during or after data collection. Stratified random sampling was applied to obtain patients from 4 groups of departments: medicine, surgery, orthopedics, and obstetrics and gynecology. Subsequently, systematic random sampling was used via a computer system to select patient records from each department to ensure equivalent representativeness. The required sample size was calculated using the formula for finite populations, targeting a 95% confidence level and 5% margin of error. An estimated population of 31,103 discharges in 2023 yielded a required sample size of 384. A 20% buffer for incomplete data was added, yielding a final target of 445 records.

This study included secondary information from the discharge summaries in the EMR system, casemix-coded patient data, and coder demographic profiles. A comparison was made between the coders’ codes and the codes reassigned by a trained medical officer and casemix experts. The gold-standard codes were determined through a structured review process conducted by a trained medical officer with casemix coding experience and verified by certified casemix experts at HCTM. Each sampled record was independently reviewed against the discharge summary in the EMR, and the reassigned code was agreed upon through consensus between the medical officer and the casemix expert. In cases of disagreement, a third senior casemix expert served as the final arbiter to ensure a consistent and predefined adjudication process. To minimize assessment bias, the trained medical officer and casemix experts who reassigned the principal diagnosis codes were blinded to the original coder-assigned codes at the time of review. The original codes were only compared after the gold-standard codes had been independently assigned, ensuring that the reassignment process was not influenced by knowledge of the coders’ prior decisions. In this study, the dependent variable was coding accuracy, defined as whether the principal diagnosis was coded correctly by the casemix coder, identical to the codes assigned by the trained medical officer or casemix experts. The data were categorized into 2 groups: accurate or inaccurate. This study deliberately focused on the accuracy of the principal diagnosis code only rather than secondary diagnoses or procedure codes. This decision was based on the fact that the principal diagnosis is the primary driver of My-DRG casemix grouping and hospital reimbursement in the Malaysian casemix system, making it the most consequential code for financial and policy outcomes. It is important to note that this study evaluated the accuracy of the International Classification of Diseases, Tenth Revision, principal diagnosis coding specifically rather than the resulting My-DRG group assignment itself. My-DRG grouping accuracy depends on the joint accuracy of multiple coded elements (principal diagnosis, secondary diagnoses, procedures, age, and length of stay) and was beyond the scope of this study. Furthermore, restricting the analysis to the principal diagnosis allowed for a focused and interpretable measure of coding performance, consistent with several prior studies in similar health care settings [7].

Four components of independent variables were included: (1) patient-related factors such as age, sex, and length of stay; (2) clinical departments; (3) documentation factor; and (4) coder-related factors, namely, length of service, educational level, and time taken to assign a clinical code. The length of stay in hospital was defined as the number of days that the patient stayed in the hospital and was further divided into inliers and outliers. The discharge summary was reviewed for completeness of diagnosis. Documentation was considered complete if an actual primary diagnosis was recorded. Specifically, a discharge summary was considered complete if it contained an explicitly stated primary (principal) diagnosis written in full clinical terminology without reliance on abbreviations, symbols, or ambiguous shorthand. Summaries that contained only nonspecific descriptors (eg, “unwell,” “fever,” or “query diagnosis”), left the primary diagnosis field blank, or contained illegible entries were classified as incomplete. This operational definition was consistently applied by the reviewing medical officer across all sampled records.

For coder-related factors, the length of service was divided into less than 10 years and greater than or equal to 10 years. This threshold was used as it is consistent with the minimum reckonable service period for confirmed or pensionable status and optional retirement eligibility under Malaysia’s Pensions Act 1980 (Act 227), section 12 [20]. Educational level was categorized as having a degree or not, and the time taken for the coding officer to complete coding was calculated from the date of discharge until the date when the coding data were finalized. This period was divided into 2 categories: 3 days and below and 4 days and above. These cutoff points are consistent with the international discharged, not final coded benchmark used in health information management departments, where 3 days after discharge is the standard threshold for flagging a record as overdue for coding completion [21].

Descriptive analysis was conducted to present the background and study variables. Bivariate analyses were performed to understand the association between independent variables and the dependent variable. All associations were considered statistically significant if the P value was below .05. All data were recorded in Microsoft Excel and analyzed using SPSS (version 29; IBM Corp). Bivariate analysis using the Pearson chi-square test was performed to examine whether the coding accuracy status (accurate vs inaccurate) differed significantly across subgroups of each patient and coder characteristic. This approach was adopted to identify which specific variables were statistically associated with coding inaccuracy, thereby allowing for the prioritization of variables for targeted quality improvement interventions. In addition to the test of association, the coding accuracy rate for each variable category was calculated as the proportion of accurately coded records out of the total records in that category to provide a more interpretable measure of coding performance at the variable level. All 445 sampled records had complete data for discharge summary completeness status and coder-related variables. No records required exclusion due to missing data.

This study received ethics approval from the Medical Research and Ethics Committee of Universiti Kebangsaan Malaysia, with approval reference number JEP-2024-289. All patient data were anonymized to ensure participant confidentiality, and no individual participants could be identified during or after data collection. Informed consent was not required as the study used deidentified retrospective administrative data.

This study aimed to assess the accuracy of clinical coding for principal diagnoses at a major Malaysian teaching hospital. The findings revealed an overall coding accuracy rate of 76.4% (340/445). These results are consistent with those of prior literature but highlight room for improvement. For example, Olagundoye et al [22] reported a slightly higher accuracy rate of 78.7% at Lagos State University Teaching Hospital. Worldwide, clinical coding accuracy varies substantially, with a systematic review by Burns et al [23] identifying a median accuracy rate of 83.2%, ranging from as low as 50.5% to as high as 97.8%. Our findings fall within this range but remain below the optimal threshold for high-performing health systems.

Interestingly, our error rate was lower than the 49.8% reported by Zafirah et al [7] and comparable to the 25.2% observed by Saizan et al [6]. This difference may be attributed to methodological differences: while our study focused solely on the principal diagnosis, many others have evaluated multiple diagnoses and procedural codes, which are inherently more complex and prone to error. Such methodological clarity is a strength of our study, allowing for a focused and interpretable measure of accuracy that reflects core hospital performance in coding.

One of the most significant findings was the strong association between the completeness of discharge summaries and coding accuracy. In our sample, 88.1% (392/445) of the records were complete, substantially higher than the 7% completeness rate reported by Zafirah et al [7]. This likely stems from our focus on the principal diagnosis, whereas previous studies have assessed full diagnostic and procedural completeness. Nevertheless, our statistical analysis confirmed that complete documentation significantly improves coding accuracy (P<.001). These results align with those of Taiwo Adeleke et al [24], who reported that incomplete or poor-quality documentation was a leading contributor to coding inaccuracy in Nigerian hospitals. Moreover, Foster et al [5] emphasize that continuous auditing and feedback, combined with electronic discharge summary systems, can meaningfully enhance both documentation and coding standards. Active clinician engagement in the documentation process, as advocated by Nouraei et al [25], is also vital as the absence of clear diagnostic notes can critically impair the coder’s ability to assign appropriate codes.

Another notable finding was the statistically significant association between coding timeliness and accuracy. Records coded within 3 days of discharge were significantly more accurate than those coded after 4 or more days (75% vs 25%; P=.02). This supports the view that shorter coding turnaround times may reflect more efficient workflows or more straightforward cases. However, longer durations could also indicate increased care and attention to complex cases. Studies have found that coders who took additional time often achieved better accuracy, particularly for complex cases [26,27]. This raises the possibility that longer turnaround time may, in part, reflect greater case complexity rather than inefficient workflow alone, a distinction this study could not disentangle as case-level complexity (eg, number of comorbidities, diagnostic ambiguity, and length and diversity of documentation) was not independently measured. Coding complexity has been recognized elsewhere as a substantive, measurable construct that affects coder workload and throughput independent of coder efficiency [26,27], and future studies should consider incorporating a complexity measure (eg, number of secondary diagnoses, comorbidity index, or casemix index) to formally test whether case complexity confounds the turnaround time–accuracy relationship. Conversely, studies have also reported that prolonged delays might signal coder fatigue, systemic inefficiencies, or excessive workloads, factors that could compromise coding quality [28,29]. These findings suggest that turnaround time is a multifaceted indicator that could be used by hospital administrators to identify bottlenecks and improve workflow efficiency.

While the analysis of department-specific trends showed that obstetrics and gynecology had the highest rate of coding errors followed by orthopedics, internal medicine, and surgery, these differences were not statistically significant. Nevertheless, this trend aligns with findings by Zafirah et al [7] indicating that certain specialties may be more vulnerable to documentation and coding errors. This calls for department-specific interventions such as targeted coder training or closer clinician-coder collaboration, particularly in departments with complex case profiles.

Interestingly, coder-related factors such as educational attainment and years of experience did not significantly affect coding accuracy in this study. While this contrasts with literature suggesting that formal qualifications and experience enhance coder performance, our findings may be due to the small number of coders (n=4) and the relatively uniform training they receive at the ITCC. Future studies with larger and more diverse coder samples could provide more definitive insights into these relationships.

This study has several strengths. It is one of the few Malaysian investigations to focus specifically on principal diagnosis coding, providing a clear and interpretable metric of performance. It also combines administrative data with manually validated recoding, enhancing data reliability. Moreover, the statistically significant findings on discharge summary completeness and coding timeliness offer actionable insights for quality improvement.

However, some limitations must be acknowledged. This study was conducted at a single teaching hospital, limiting generalizability to other health care settings, including private or nonacademic hospitals. Teaching hospitals such as HCTM typically have access to specialist input; structured coder training programs; and more complex, tertiary-level case mixes than nonteaching public hospitals. Nevertheless, teaching hospitals may also differ from private hospitals in EMR infrastructure, documentation culture, and staffing ratios of coders to caseload. The coding accuracy patterns observed in this study, particularly the influence of documentation completeness and turnaround time, may therefore not directly generalize to district or state public hospitals or private health care settings, and multisite studies are recommended to confirm these findings across different hospital types.

Additionally, the small number of coders limited our ability to explore demographic factors robustly. With only 4 coders contributing to the 445 coded records, coder-level variables (such as educational level and years of service) were effectively compared across just 4 clusters rather than 445 independent units, substantially limiting statistical power to detect true coder-level effects. Nonsignificant associations for these variables should therefore be interpreted with caution rather than as confirmed null findings. This study also did not engage clinicians directly to understand root causes of documentation lapses, which could have enriched the analysis. Additionally, a formal interrater reliability assessment between the trained medical officer and casemix experts was not conducted, which is acknowledged as a limitation of this study. Future studies should incorporate a structured reliability assessment such as the Cohen κ to strengthen the validity of the gold-standard coding process.

It is acknowledged that the use of separate bivariate chi-square tests to examine associations between individual variables and coding accuracy status has limitations as this approach does not account for confounding between variables. In particular, potential confounders such as comorbidity burden, admission type (elective vs emergency), discharge disposition, and overall casemix severity were not captured in this dataset and could not be adjusted for. As this was a preliminary descriptive study designed to characterize bivariate associations rather than isolate independent predictors, these factors represent important variables for inclusion in future multivariable analyses. Future studies should consider using multivariable logistic regression, with coding accuracy status as the dependent variable and coder-related factors, including educational level, years of service, time taken for coding, and department, as independent variables. Such an approach would yield adjusted estimates and provide more robust evidence on the independent predictors of coding inaccuracy. In this analysis, chi-square tests were used to examine bivariate associations between coding accuracy and the variables of interest, rather than to identify independent predictors. The findings were intended to describe patterns of association and inform variables that may warrant further investigation in future multivariable analyses.

The analysis was limited to bivariate associations as an exploratory first step. A multivariable logistic regression model was not developed as this study aimed to describe preliminary associations rather than establish causal or predictive relationships. The small number of coders (n=4) further precluded robust multivariable modeling, and future studies with larger coder samples are encouraged to build on these findings. A multivariable logistic regression model was therefore not pursued at this stage as the primary aim was to describe the pattern of associations across variables rather than determine independent predictors. Future studies with larger and more diverse samples are recommended to perform logistic regression analysis to confirm these preliminary findings and identify independent predictors of coding inaccuracy.

Additionally, a sensitivity analysis restricting the sample to records with complete discharge summaries was not performed in this preliminary study. Given that documentation completeness was the strongest predictor of accuracy, future work should also examine whether other associations (eg, turnaround time and years of service) persist when analysis is restricted to records with complete documentation to rule out confounding by documentation completeness itself.

Despite these limitations, the findings have important implications for policymakers and hospital administrators. Improving the completeness and clarity of discharge summaries through training, digital templates, and clinical documentation improvement initiatives can directly enhance coding accuracy. In turn, more accurate coding enables better casemix classification, fairer hospital reimbursements, and more reliable health service planning. The significant relationship between coding timeliness and accuracy also suggests the need for balanced workload distribution and process optimization to reduce coding delays without compromising quality.

Hospitals should implement a structured, recurring coding audit program, for example, auditing 3.5% to 5% of monthly coded records on a weekly cycle, consistent with established health information management auditing practice, rather than relying on ad hoc or annual reviews [30]. Coder feedback should be delivered promptly after each audit cycle, with individualized scorecards highlighting recurring error patterns (eg, specific diagnosis categories or documentation gaps). A formal, structured clinician query process following established frameworks such as the American Health Information Management Association and Association of Clinical Documentation Integrity Specialists Guidelines for Achieving a Compliant Query Practice should be adopted to standardize how coders seek clarification from clinicians when discharge documentation is ambiguous or incomplete [30]. Standardized discharge summary templates with mandatory fields for principal diagnosis and key clinical indicators should also be piloted to reduce documentation-related coding errors, the single strongest predictor of inaccuracy identified in this study.

As health systems move toward performance-based financing and data-driven policy, coding accuracy becomes not merely a technical concern but also a strategic priority. In this context, this study’s findings offer a timely reminder that investments in documentation quality, coder support, and integrated clinical workflows are essential to realizing the full value of casemix systems in Malaysia and beyond.

The results of this study showed that clinical coding accuracy was 76.4% (340/445), lower than the key performance index set by the Ministry of Health of Malaysia, which is 90%. The findings underscore the importance of complete documentation and timely coding in enhancing the accuracy of clinical coding in casemix systems. Hospitals should invest in ongoing coder training and implement systematic audits to ensure high-quality data, which is critical for planning, financing, and evaluating health care services.