External Validation and Adaptation of the Cambridge Multimorbidity Score Using German Claims Data

ABSTRACT

Objective: The Cambridge Multimorbidity Score (CMMS) is a well-regarded indicator for predicting primary care visits, hospital admissions, and mortality. However, it is not compatible with the International Classification of Diseases, Tenth Revision (ICD-10). This study aimed to translate the CMMS to ICD-10, assess its performance, and evaluate its usefulness in primary care using German claims data. We then compared 3 measures—our translated ICD-10 version of the CMMS, the Charlson Comorbidity Index, and a version recalibrated to our study population to predict hospitalizations, mortality, and ambulatory care–sensitive conditions (ACSCs).

Study Design: Retrospective observational cohort study (2011-2022). Two data sets were used: a whole adult insured population (3,809,220) with 1-year follow-up, and a subset of 217,964 patients with diabetes mellitus with 11-year follow-up.

Methods: The predictive performance for binary outcomes—hospitalization, mortality, and ACSCs—was assessed through external validation of the CMMS, using the original Payne et al score. In addition, we re-estimated the weights for the existing conditions from the original CMMS to reflect their prevalence in the German population.

Results: The CMMS outperformed the CCI by 2.7 percentage points for ACSCs, 3.6 percentage points for mortality, and 4.0 percentage points for hospitalization within 1 year. Over 11 years, Payne’s CMMS showed the highest discrimination (C-index 0.663 vs 0.651 for CCI). The CMMS with population-derived weights performed similarly to the original CMMS in all analyses.

Conclusions: The CMMS demonstrated improved predictive accuracy over the CCI for mortality, hospitalizations, and ACSCs, supporting its use for risk stratification and health care planning in German primary care and comparable populations.

Am J Manag Care. 2026;32(10):e372-e378

doi:10.37765/ajmc.2026.90033

Takeaway Points

• The Cambridge Multimorbidity Score (CMMS) was successfully adapted to International Classification of Diseases, Tenth Revision, German Modification for use with German statutory health insurance claims data.
• The International Classification of Diseases, Tenth Revision–based CMMS demonstrated better predictive performance than the Charlson Comorbidity Index (CCI) for hospitalizations, mortality, and ambulatory care–sensitive conditions (ACSCs).
• The CMMS showed consistent predictive performance across both short-term and long-term follow-up periods.
• Population-derived CMMS weights performed similarly to the original weights developed by Payne et al, supporting the robustness of the original score when applied to German claims data.
• The adapted CMMS may support risk stratification, identification of patients with multimorbidity, and health care planning in German primary care and comparable health care settings.

In epidemiological research focusing on health outcomes, it is often essential to describe both morbidity and comorbidity within patient groups or across broader populations.^1-3 These descriptions are valuable in analyses aimed at examining outcomes such as mortality, treatment efficacy, health care utilization, or health care costs.^1,4 Over recent decades, numerous morbidity indices have been developed to address specific analytical needs.⁴ Among these, the Charlson Comorbidity Index (CCI) has become the most widely used tool.^1,5 Developed in 1987, this index was designed to account for comorbid conditions that could affect mortality among patients admitted to the medical service at a New York hospital.^1,4 There have also been several modifications to the CCI, including different selections of diagnoses and varying weights assigned to them.^6-9 However, the relevance of the CCI is being questioned, as it includes only 19 comorbid diseases and overlooks important mortality-related health indicators identified in recent years.^10,11 Furthermore, the approach to managing multimorbidity has shifted toward a greater emphasis on primary care and on managing diseases outside hospital settings.^3,12 The CCI was designed for use with hospitalized patients and may not be ideally suited for research in primary care or public health settings.⁴ The absence of dementia and psychiatric diagnoses in this index, potential allocation of an excessive number of points for patients with HIV, and a rather poor age adjustment are also important limitations of the CCI.^4,13,14 As a result, there is a need for morbidity indices developed specifically for primary care or public health settings.

To address the above-described limitations, Payne et al¹⁵ developed and validated the Cambridge Multimorbidity Score (CMMS) for use in primary care practices, utilizing data from the Clinical Practice Research Datalink (CPRD). The CMMS incorporates 20 diseases to predict primary care consultations, unplanned hospital admissions, and mortality.¹⁵ The CMMS provides reliable, consistent estimates and is widely adaptable for use in various medical environments.¹² Furthermore, it helps pinpoint high-risk patients, enabling more strategic and effective health care planning.¹⁵

The databases used for the CMMS predominantly originate from primary care settings in the United Kingdom, where diagnoses are encoded using terminologies such as SNOMED CT or, previously, Read codes. SNOMED CT is a comprehensive, hierarchically organized clinical terminology that offers high granularity and interoperability, enabling detailed and specific documentation of clinical concepts. Although the original CMMS was developed using SNOMED CT and Read codes, the underlying disease definitions are universal. We adapted these definitions to the International Classification of Diseases, Tenth Revision (ICD-10), specifically the International Classification of Diseases, Tenth Revision, German Modification (ICD-10-GM), used for diagnosis coding in German statutory health insurance claims data. ICD-10 is an internationally standardized classification system that supports epidemiological research and cross-country data comparisons.^12,16

The aim of this study was to perform an external validation of the CMMS using ICD-10–coded German primary care claims data. This approach allows us to evaluate the predictive performance of the original CMMS relative to outcomes including hospitalizations, mortality, and ambulatory care–sensitive conditions (ACSCs). In addition, we reestimated the weights for the existing conditions from the original CMMS to reflect their prevalence in the German population.

METHODS

Data Protection Statement

The study used pseudonymized statutory health insurance claims data provided by Allgemeine Ortskrankenkasse (AOK) Baden-Württemberg. Data were analyzed in accordance with applicable data protection regulations, including the General Data Protection Regulation, and no directly identifiable personal information was accessible to the researchers.

Setting and Study Design

We conducted a retrospective cohort study using claims data from one of the largest statutory health insurance providers in Germany, the AOK in the federal state of Baden-Württemberg, from 2011 to 2022. The study included 3,809,220 insured individuals who were members of AOK Baden-Württemberg, resided in Baden-Württemberg, and were aged 20 to 95 years.

To fulfill the study’s aim, the estimates were obtained in 2 ways. The first part assessed the relative performance of the CMMS in predicting hospitalizations, all-cause mortality, and ACSCs across all patients. The second was to evaluate the CMMS’s ability to predict patient survival for a predefined health condition; we selected patients with diabetes. Both steps were performed relative to the CCI. For survival time analysis, a subgroup of patients was selected from data collected between 2011 and 2022, comprising a closed cohort of 217,964 patients diagnosed with diabetes (referred to here as Data Set 1). This data set represents a sicker population with frequent health care utilization, allowing longitudinal evaluation of the CMMS in a more high-risk subgroup. The original CMMS for binary outcomes was used, which is a second data set from the year 2021 (further referred to as Data Set 2) comprising 1,949,518 insured individuals. This data set provides a snapshot of the broader insured population actively using primary care services, allowing assessment of the CMMS performance in a general population.

Outcomes

Binary outcomes (coded as 1 – “yes,” 0 – “no”) were analyzed over the observation period as follows:

All-cause hospitalization (January 1, 2022-December 31, 2022): At least 1 hospitalization, including readmissions, with at least 1 overnight stay during the year under observation

All-cause mortality (January 1, 2022-December 31, 2022): Mortality representing all-cause mortality within the observation period. All of these logistic regression outcomes were recorded annually.

ACSCs (January 1, 2022-December 31, 2022): ACSCs were computed according to the Purdy ACSC definition and using ICD-10 codes.¹⁷ This dichotomous outcome was set to 1 if the patient had at least 1 ACSC-indicated hospitalization stay per year.

Adaptation of the CMMS

Our primary objective was to evaluate whether the CMMS developed by Payne can be applied reliably in a different health care system, using a distinct ICD coding structure and a broader patient population. We consider the use of internationally established morbidity scores essential for enabling meaningful comparisons of outcomes across countries and health systems. As part of our evaluation, we also derived condition weights from our own data—using the same original conditions listed in Table 1¹⁵—to examine whether a locally calibrated version of the CMMS might offer any improvement. This comparison was conducted solely to assess potential differences; the weighting scheme proposed by Payne remains our primary reference.

The CMMS was applied using ICD-10-GM coding to enable its evaluation within the German health care system. This study performs an external validation of the original CMMS using the existing weights assigned to 20 chronic conditions. The complete list of diagnoses and their corresponding weights from the original CMMS is provided in Table 1.

Statistical Analysis

Comorbidity Scores

Three comorbidity measures were evaluated: the CCI, the original CMMS proposed by Payne, and a recalibrated CMMS version with weights derived from our study population (CMMS adjusted).

Data Sets

Two data sets were used to evaluate the performance of the comorbidity scores. The first data set (Data Set 1) consisted of a diabetic subgroup (n = 217,964) in which the scores were estimated in 2011, and individuals were followed until 2022, allowing long‑term performance assessment for 11 years. The second data set (Data Set 2) comprised the entire population aged 20 to 95 years (n = 3,809,220), for whom all 3 scores were calculated in 2021 and linked to a 1‑year follow‑up period. This dual‑cohort approach enabled evaluation of CMMS performance in both the general population and a clinically relevant high‑risk subgroup.

Recalculating New Weight for the CMMS (Adjusted)

To derive the CMMS adjusted score, we followed an approach conceptually similar to the one described by Payne. First, separate logistic regression models were fitted for each of the three 1‑year outcomes—ACSC events, mortality, and hospitalization—using the full patient cohort. Each model included a full set of 20 comorbid conditions used in Payne’s original CMMS formulation. From these models, we extracted the regression coefficients and their standard errors. To obtain a single, stable weight for each comorbidity, the condition‑specific estimates were pooled across the 3 outcomes, using an inverse‑variance weighting approach. The resulting pooled estimates served as the weights in the construction of the CMMS adjusted score. A complete table of the condition‑specific estimates from all 3 outcome models, along with the pooled values used in the final weighting scheme, is provided in the eAppendix (eAppendix available at ajmc.com).

Survival Analysis

For each score, quartiles were calculated based on the distribution of values across the cohort, including all patients. Associations between comorbidity quartiles and all‑cause mortality were assessed using Cox proportional hazards models. For each score, we estimated HRs with 95% CIs across quartiles, using the lowest quartile as the reference. Models were run both unadjusted and adjusted for age and sex. The Cox model using the continuous score values did not satisfy the proportional hazards assumption. After categorizing the score into quartiles, all proportional hazards assumptions were met. These assumptions were evaluated using Schoenfeld residual plots. The survival analyses were performed using the function from the survival package in R (R Foundation for Statistical Computing R Version 4.5.1, packages: survival [version 3.8-3], survminer [version 0.5.0]).

Discrimination Analysis

We quantified discrimination using the area under the receiver operating characteristic curve (AUC-ROC). AUC was calculated in 2 ways: first, using ROC curves for 1‑year follow‑up in the entire patient cohort, and second, using time‑dependent survival analysis over 11 years, restricted to patients with diabetes.

For each outcome–score combination, predicted probabilities were generated from logistic regression models with a binomial distribution, and ROC curves were constructed using the pROC package version 1.19.0.1 and timeROC version 0.40 in R. AUC values were extracted directly from these ROC objects, and 95% CIs were estimated using the DeLong method, providing a robust nonparametric measure of uncertainty. To formally compare the discriminatory performance of the morbidity scores, we performed pairwise DeLong tests. These comparisons were based on ROC curves derived from the raw score values, enabling direct evaluation of whether CMMS Payne differed significantly from CCI or from CMMS adjusted. For each comparison, we reported the AUC difference, its 95% CI, and the corresponding P value. All ROC and calibration curves (unadjusted and with adjustment for age and sex) generated as part of this analysis are presented in the eAppendix.
Time‑dependent predictive performance was assessed using measures derived from the Cox proportional hazards models. For each comorbidity score, discrimination over time was quantified using time‑dependent AUCs at 1, 5, and 10 years, calculated from model‑based survival estimates. Overall discrimination was evaluated using the Harrell C‑index, and prediction accuracy was assessed using the time‑dependent Brier score at the same intervals.

RESULTS

Survival Analysis

Across all 3 comorbidity measures, mortality risk increased steadily with each higher quartile. However, the strength of association and the discriminative performance varied between the scores (Table 2). In the unadjusted models, the CCI showed a strong gradient, with HRs rising from 1.61 in quartile 2 (Q2) to more than 4.0 in Q4. The CMMS Payne score exhibited a very similar pattern, but with slightly higher HRs in the upper quartiles, indicating a somewhat stronger separation between risk groups. The CMMS adjusted score, recalibrated for the study population, yielded HRs comparable to those from CMMS Payne, although slightly lower in the highest quartile. Discrimination was modest for all 3 unadjusted models, with C‑indices ranging from 0.644 (CCI) to 0.653 (CMMS Payne), suggesting that CMMS Payne provided the best unadjusted predictive performance.

After adjustment for sex and age, HRs decreased in magnitude across all scores, as expected, but the relative ordering of model performance remained consistent. CCI showed attenuated but still substantial risk gradients (HR, 1.32 in Q2 to 2.48 in Q4). Both CMMS Payne and CMMS adjusted retained stronger associations than CCI, with adjusted HRs of approximately 1.18 to 2.53 and 1.19 to 2.49 across quartiles, respectively. Importantly, adjustment markedly improved discrimination for all models. The C‑index increased to 0.758 for CCI and to 0.759 for both CMMS versions, indicating that once demographic factors were accounted for, the CMMS Payne and CMMS adjusted scores performed nearly identically and slightly better than CCI.

Overall, the results show that all 3 scores effectively stratify mortality risk, but the CMMS‑based scores consistently outperform CCI, particularly after adjustment. The recalibrated CMMS adjusted score performs on par with the original CMMS Payne, suggesting that the weighting scheme derived for our population preserves predictive ability while aligning the score more closely with local data.

Time‑Dependent Predictive Performance

Across all 3 comorbidity scores, discrimination was highest at the 1‑year time point and declined gradually over longer follow‑up (Table 3). In the unadjusted models, the CMMS Payne score consistently performed best. At 1 year, its AUC was 0.781, slightly higher than those of both the CCI (0.771) and the CMMS adjusted scores (0.769). This advantage persisted at 5 and 10 years, where CMMS Payne maintained the highest AUC values (0.724 and 0.701, respectively). The C‑index values followed a similar pattern, with CMMS Payne achieving the best overall discrimination (0.663) compared with 0.651 for CCI and 0.657 for the adjusted CMMS. Brier scores were nearly identical across all 3 models at 1 year, but diverged slightly at longer horizons, with CMMS Payne showing marginally better calibration at 5 and 10 years.

After adjustment for age and sex, all 3 scores showed substantial improvement in discrimination. The C‑index increased to approximately 0.762 to 0.763 for all models, indicating that demographic adjustment largely equalized performance between the scores. AUC values at 1, 5, and 10 years were also nearly identical across CCI, CMMS Payne, and CMMS adjusted, with all 3 achieving AUCs of approximately 0.80 at 1 year and 0.825 at 10 years. Brier scores also improved in the adjusted models, particularly at longer time points, reflecting better overall predictive accuracy.

In summary, these results show that CMMS Payne provides the strongest discrimination in unadjusted analyses, outperforming both CCI and the recalibrated CMMS adjusted score. However, once age and sex are incorporated, all 3 scores converge to nearly identical predictive performance, with very similar AUCs, C‑indices, and Brier values across all time horizons. This suggests that demographic factors account for much of the difference between the scores, and that the CMMS adjusted score performs on par with the original CMMS Payne when evaluated in a fully adjusted setting.

Predictive Performance Across Outcomes Based on 1-Year AUC Estimates

Across all 3 outcomes evaluated at 1‑year follow‑up, the CMMS‑based morbidity scores consistently outperformed the CCI in discrimination, calibration, and overall predictive accuracy (Table 4). For ACSC events, the CMMS adjusted score showed the strongest discrimination (AUC, 0.803; 95% CI, 0.800-0.805), followed closely by CMMS Payne (AUC, 0.799), whereas CCI performed notably worse (AUC, 0.772). Calibration intercepts were near zero for all models, and Brier scores were uniformly low (0.009), indicating good prediction accuracy. The AUC for CMMS Payne was 2.7 percentage points higher than that for CCI, indicating modestly better discrimination for identifying patients who would experience an ACSC event within 1 year.

For 1‑year mortality, all models performed strongly, but the CMMS again provided superior discrimination. CMMS Payne achieved the highest AUC (0.866; 95% CI, 0.865-0.868), followed closely by CMMS adjusted (0.864), whereas CCI lagged (0.830). Brier scores were low (0.013-0.014) and calibration remained excellent. The AUC difference between CMMS Payne and CCI was 0.036, indicating that CMMS Payne was 3.6 percentage points better at correctly distinguishing who would die within 1 year and who would not.

Prediction of 1‑year hospitalization was more challenging overall, reflected in lower AUCs and higher Brier scores. Even so, the CMMS adjusted score performed best (AUC, 0.716; 95% CI, 0.715-0.717), followed by CMMS Payne (AUC, 0.712), both of which outperformed CCI (AUC, 0.673). The AUC difference of 0.040 between CMMS Payne and CCI indicates a 4‑percentage‑point improvement in discrimination.

When models were adjusted for age and sex, performance improved across all outcomes. CMMS adjusted remained the strongest predictor for ACSC and hospitalization, while CMMS Payne and CMMS adjusted performed nearly identically for mortality. Across all analyses, the CMMS demonstrated superior discrimination, excellent calibration, and lower prediction error than CCI, confirming its stronger predictive value for 1‑year outcomes.

DISCUSSION

The results of this study demonstrate that the adapted CMMS outperforms the CCI in several key areas, including mortality, all-cause hospitalizations, and ACSCs. These results are consistent with other authors’ findings for the CMMS.¹⁵ Across the unadjusted models, the CMMS consistently achieved higher predictive performance, with significantly higher AUC values than the CCI. This indicates that the CMMS may offer a more accurate estimation of hospitalization risk, which is crucial for health care planning, targeted prevention, and efficient resource allocation. Together with other authors, we also advocate for the use of CMMS, as its modified version enables reliable estimation using clinical terminology that is universally applicable across various health care environments.¹²

ACSCs are often interpreted as indicators of suboptimal outpatient care or disease management; the ability of the CMMS to better predict these events supports its value in identifying patients at risk of potentially avoidable hospitalizations and in informing targeted intervention strategies. Health costs due to ACSC-related hospitalization are enormous, reaching billions of dollars annually.¹⁸ Furthermore, the number of ACSC diagnoses continues to grow, and the costs of treating these conditions continue to rise.¹⁹ In many cases, hospitalizations could be avoided, yielding benefits for both the health care system as a whole and individual patients.¹⁹ This indicates a great need for more precise measures of ACSC hospitalization rates.

Mortality increased consistently with higher comorbidity burden, and all 3 scores demonstrated clear and stable risk stratification across both the general population and the diabetic subgroup. Hospitalization risk showed a similar pattern, with higher score levels associated with progressively greater likelihood of admission, underscoring the clinical relevance of comorbidity assessment for predicting both fatal and nonfatal outcomes. In both analyses, CMMS outperformed the CCI, providing sharper discrimination for mortality and hospitalization and offering a more informative risk gradient across score levels. The superior performance of CMMS compared with the CCI likely reflects its broader and more contemporary representation of chronic disease burden. CMMS incorporates a wider range of conditions and uses empirically derived weights that more accurately mirror real‑world risk, resulting in clearer gradients for both mortality and hospitalization.

A slightly lower AUC in the diabetic subgroup compared with the full population suggests that model performance may vary across clinical subgroups, even within the same country. This pattern indicates that differences in underlying morbidity profiles could influence how well a comorbidity score performs. At the same time, the CMMS adjusted score performs almost identically to the original CMMS Payne, which implies that national‑level population differences may play a smaller role than differences between specific clinical subgroups. However, confirming whether subgroup‑specific recalibration is necessary would require additional targeted analyses.

CMMS consistently outperformed the CCI for both mortality and hospitalization, showing clearer risk stratification and stronger discrimination across score levels. These findings indicate that CMMS captures clinically relevant comorbidity patterns more effectively than the CCI, particularly in predicting both fatal and nonfatal outcomes. The better results obtained by the adapted CMMS compared with CCI stem from the CMMS index being based on updated comorbidity indices, which often improve risk stratification by incorporating more recent data.

Strengths and Limitations

This study has several notable strengths, including the use of a large, contemporary German claims database, the systematic adaptation of the CMMS to the German ICD-10-GM coding system, and external validation of the score in a German population. The study further provides both short- and long-term performance assessments and compares the original CMMS with the established CCI and a population-recalibrated version. However, there are also some limitations to consider. The use of ICD-10-GM codes, while appropriate for this context, is less detailed than databases such as CPRD, so the translation of clinical concepts may not be entirely precise. Nonetheless, the results align well with those of other studies, suggesting that the models are robust and well-calibrated. Furthermore, diagnostic coding in medical records is primarily performed for clinical, not research, purposes, which can lead to potential misclassification or missing data. However, the scores still reflect realistic outcomes expected in everyday clinical practice.

The choice between CMMS and CCI may ultimately depend on the specific clinical outcome of interest. It is important to note that the CCI and CMMS serve different purposes. Although the CCI was originally developed to predict mortality and comorbidity burden, the CMMS is designed for use in primary care settings to predict outcomes related to general practitioner utilization and other primary care measures. Therefore, this study was not intended to establish a “better” score overall, but rather to provide a viable alternative to the CCI for use in primary care contexts. Further research and validation in diverse populations are warranted to confirm these findings and to evaluate the practical benefits of implementing the CMMS in routine care.

CONCLUSIONS

The CMMS offers improved predictive accuracy over the CCI for mortality, hospitalizations, and ACSCs, making it a promising tool for risk stratification, chronic disease management, and health care planning in primary care. These findings indicate that CMMS performs robustly and can be reliably applied in the German population and, likely, in other comparable populations as well.

Author Affiliations: Goethe University Frankfurt, Faculty of Medicine, Institute of General Practice (KK, CF, AG, JWW), Frankfurt am Main, Germany; Goethe University Frankfurt, Faculty of Medicine, Institute of Biostatistics and Mathematical Modelling (ID), Frankfurt am Main, Germany; Goethe University Frankfurt, Faculty of Mathematics, Institute of Mathematics (ID), Frankfurt am Main, Germany

Source of Funding: None

Author Disclosures: The authors report no relationship or financial interest with any entity that would pose a conflict of interest with the subject matter of this article.

Authorship Information: Concept and design (KK, CF); acquisition of data (KK, JWW); analysis and interpretation of data (KK, ID, AG, JWW); drafting of the manuscript (KK, CF); critical revision of the manuscript for important intellectual content (KK, ID, CF, JWW); statistical analysis (AG, JWW); provision of patients or study materials (KK, AG); administrative, technical, or logistic support (KK); and supervision (KK).

Address Correspondence to: Kateryna Karimova, Goethe University, Theodor-Stern-Kai, 7 Frankfurt, Germany 60590. Email: karimova@allgemeinmedizin.uni-frankfurt.de

REFERENCES

1. Yurkovich M, Avina-Zubieta JA, Thomas J, Gorenchtein M, Lacaille D. A systematic review identifies valid comorbidity indices derived from administrative health data. J Clin Epidemiol. 2015;68(1):3-14. doi:10.1016/j.jclinepi.2014.09.010

2.Perkins AJ, Kroenke K, Unützer J, et al. Common comorbidity scales were similar in their ability to predict health care costs and mortality. J Clin Epidemiol. 2004;57(10):1040-1048. doi:10.1016/j.jclinepi.2004.03.002

3. Schneeweiss S, Maclure M. Use of comorbidity scores for control of confounding in studies using administrative databases. Int J Epidemiol. 2000;29(5):891-898. doi:10.1093/ije/29.5.891

4. Sandvik H, Ruths S, Hunskaar S, Blinkenberg J, Hetlevik Ø. Construction and validation of a morbidity index based on the International Classification of Primary Care. Scand J Prim Health Care. 2022;40(2):305-312. doi:10.1080/02813432.2022.2097617

5. Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying prognostic comorbidity in longitudinal studies: development and validation. J Chronic Dis. 1987;40(5):373-383. doi:10.1016/0021-9681(87)90171-8

6. D’Hoore W, Sicotte C, Tilquin C. Risk adjustment in outcome assessment: the Charlson Comorbidity Index. Methods Inf Med. 1993;32(5):382-387.

7. Ghali WA, Hall RE, Rosen AK, Ash AS, Moskowitz MA. Searching for an improved clinical comorbidity index for use with ICD-9-CM administrative data. J Clin Epidemiol. 1996;49(3):273-278. doi:10.1016/0895-4356(95)00564-1

8. Khan NF, Perera R, Harper S, Rose PW. Adaptation and validation of the Charlson index for Read/OXMIS coded databases. BMC Fam Pract. 2010;11:1. doi:10.1186/1471-2296-11-1

9. Quan H, Li B, Couris CM, et al. Updating and validating the Charlson Comorbidity Index and score for risk adjustment in hospital discharge abstracts using data from 6 countries. Am J Epidemiol. 2011;173(6):676-682. doi:10.1093/aje/kwq433

10. Kabboord AD, Godfrey D, Gordon AL, et al. The modified functional comorbidity index performed better than the Charlson index and original functional comorbidity index in predicting functional outcome in geriatric rehabilitation: a prospective observational study. BMC Geriatr. 2020;20(1):114. doi:10.1186/s12877-020-1498-z

11. Niewin´ski G, Graczyn´ska A, Morawiec S, et al. Renaissance of modified Charlson Comorbidity Index in prediction of short- and long-term survival after liver transplantation? Med Sci Monit. 2019;25:4521-4526. doi:10.12659/MSM.914669

12. Tsang RS, Joy M, Whitaker H, et al. Development of a modified Cambridge Multimorbidity Score for use with SNOMED CT: an observational English primary care sentinel network study. Br J Gen Pract. 2023;73(731):e435-e442. doi:10.3399/BJGP.2022.0235

13. Roffman CE, Buchanan J, Allison GT. Charlson Comorbidities Index. J Physiother. 2016;62(3):171. doi:10.1016/j.jphys.2016.05.008

14. Zavascki AP, Fuchs SC. The need for reappraisal of AIDS score weight of Charlson Comorbidity Index. J Clin Epidemiol. 2007;60(9):867-868. doi:10.1016/j.jclinepi.2006.11.004

15. Payne RA, Mendonca SC, Elliott MN, et al. Development and validation of the Cambridge Multimorbidity Score. CMAJ. 2020;192(5):E107-E114. doi:10.1503/cmaj.190757

16. World Health Organization. International Statistical Classification of Diseases and Related Health Problems (ICD). Access date October 19, 2026.https://www.who.int/standards/classifications/classification-of-disease

17. Purdy S, Griffin T, Salisbury C, Scharp D. Ambulatory care sensitive conditions: terminology and disease coding need to be more specific to aid policy makers and clinicians. Public Health. 2009;123(2):169-173. doi:10.1016/j.puhe.2008.11.001

18. Chen HF, Lin HR. Social determinants of ambulatory care sensitive conditions: a qualitative meta-synthesis based on patient perspectives. Front Public Health. 2023;11:1147732. doi:10.3389/fpubh.2023.1147732

19. Lui CK, Wallace SP. A common denominator: calculating hospitalization rates for ambulatory care-sensitive conditions in California. Prev Chronic Dis. 2011;8(5):A102.