How do Patient Comorbidities Influence Adult Cochlear Implant Outcomes?

Abstract

Objective

To evaluate the relationship between weighted comorbidity indices and post‐cochlear implant (CI) speech recognition and patient‐reported outcome measures (PROMs).

Study Design

Retrospective cohort.

Setting

Tertiary Neurotology Center.

Methods

Adults with bilateral hearing loss who received CIs from 2017 to 2022 were included. Comorbidities were extracted via chart review of anesthesia notes, medication lists, and clinical encounters. Three comorbidity scores were calculated: the Charlson Index (10‐year mortality), Elixhauser Index (30‐day hospital readmission), and a basic sum index assigning equal weight to 31 conditions. Spearman correlations evaluated bivariate associations. Multivariable linear regression was used to evaluate independent contributions of each index to CI outcomes, adjusting for clinical covariates. Early (1‐ or 3‐month post‐CI) and long‐term (6‐ or 12‐month post‐CI) outcomes included Consonant‐Nucleus‐Consonant word recognition (CNC) and AzBio sentence recognition in quiet (AzBioQ), and PROMs.

Results

346 patients were included. The Charlson Index was negatively associated with both early and long‐term speech recognition outcomes (CNC and AzBioQ; r _s = –0.16 to –0.31, all FDR‐adjusted P < .05). The Elixhauser Index negatively correlated with early AzBioQ scores (r _s= –0.22), while the basic sum index negatively correlated with early CNC and long‐term AzBioQ. No indices were associated with PROMs. In multivariable models, only the Elixhauser Index independently predicted early (β = –1.03, P = .005) and long‐term (β = –0.65, P = .050) AzBioQ.

Conclusion

Our findings highlight the influence of cumulative comorbid disease burden on speech recognition performance and suggests a differential impact of individual comorbidities on CI outcomes.

Hearing loss (HL) is highly prevalent, affecting 65% of adults over 71 and over 90% of those older than 90 years. ¹ Its impact extends beyond auditory deficits, contributing to fatigue, social isolation, and cognitive decline. ² Cochlear implants (CIs) provide effective auditory rehabilitation for adults with moderate‐to‐profound HL, ³ yet outcomes vary widely. ⁴ , ⁵ Known demographic, audiologic, and surgical factors explain less than half of this variability, ⁶ , ⁷ , ⁸ , ⁹ and different predictors may influence early versus long‐term outcomes. ¹⁰ Thus, identifying additional factors is critical for counseling and optimizing patient care.

One such factor is comorbidity burden, which strongly influences outcomes across chronic conditions. For example, diabetes, pulmonary disease, and chronic kidney disease worsen mortality and readmission in heart failure. ¹¹ Additionally, research has shown that each additional comorbidity is associated with reduced quality of life (QoL) in adults with chronic medical conditions. ¹² HL often coexists with chronic conditions such as diabetes and cognitive impairment, ¹³ , ¹⁴ suggesting cumulative disease burden may partly explain CI outcome variability.

In the broader audiologic literature, comorbidities have been associated with reduced hearing‐aid outcomes, with conditions such as hypertension and diabetes negatively impacting uptake and user satisfaction. ¹⁵ However, the reported relationship between comorbidities and CI outcomes has been inconsistent. For example, Aylward et al. found that an 11‐item modified frailty index (an index of physiological vulnerability) was associated with QoL (measured with the Nijmegen Cochlear Implant Questionnaire ¹⁶ ), but not speech recognition outcomes. ¹⁷ In contrast, Dang et al. reported significant differences in speech recognition outcomes between those classified as having severe comorbidities and those without comorbidities using the Adult Comorbidity Evaluation‐27 (ACE‐27, a categorical measure of comorbidity presence and severity). ¹⁸

While these studies have been informative, they are limited by the use of categorical or narrow scoring systems, potentially accounting for discrepancies in findings. The modified frailty index focuses primarily on physiologic reserve and functional status, assigning equal weight to 11 comorbidities. As such it may lack nuance in differentiating the relative impact of specific chronic conditions. The ACE‐27 evaluates 27 comorbidities assigning a grade (1‐3) for the severity of each comorbidity. However, the final score categorizes patients into one of 4 global comorbidity categories and treats comorbidities with the same severity grade as having equal impact, which may mask the differential effects of specific conditions.

In contrast, weighted comorbidity indices such as the Charlson Index and the Elixhauser Index assign differential weights to a broader range of chronic diseases, based on their relative impact on clinical outcomes. These indices were created and validated in diverse clinical populations to reflect long‐term outcomes, such as 10‐year mortality (Charlson ¹⁹ ) and 30‐day hospital readmission (Elixhauser ²⁰ ). These indices have also been used to predict a variety of chronic condition outcomes such as complications in hospitalized patients with COVID‐19, ²¹ prognosis in acute coronary syndrome, ²² and outcomes following percutaneous coronary intervention. ²³ Despite their widespread application, these tools have yet to be fully investigated in the context of adult cochlear implantation. Although originally designed to predict mortality and hospital readmission, these indices serve as proxies for overall disease burden and may also relate to functional outcomes among CI users.

To address these limitations, our study employs the Charlson Index, Elixhauser Index, and a basic sum index (which provides equal weight to each comorbidity) to evaluate how comorbidities impact postoperative (post‐CI) outcomes, including both measures of speech recognition and patient‐reported outcome measures (PROMs). We first predicted that overall greater comorbidity burden would be associated with worse CI outcomes. We further predicted that the Charlson Index, which was created to reflect more long‐term burden than the Elixhauser (10‐year mortality vs 30‐day hospital readmission), would better overlap with the factors that predict CI success, resulting in stronger relationships between CI outcomes and the Charlson Index.

Methods

Participants

The study was approved under Vanderbilt University Medical Center IRB #240876. A retrospective chart review of our clinical CI database from 2017 to 2022 was performed on 1197 total patients, identifying 346 participants with pertinent variables for inclusion. These variables included age at implantation (age), pure‐tone average (PTA) across 0.5, 1, and 2 kHz in the ear to be implanted, duration of deafness (DoD), daily CI usage (measured via datalogging), performance on speech recognition tasks, and performance on PROMs including the SSQ and CIQOL. The DoD was defined as the time since significant hearing‐aid difficulty or a notable hearing change (eg, sudden sensorineural HL). Daily CI usage was defined as mean daily processor hours from datalogging. These definitions of DoD and CI device usage were consistent with the methodology described by Defreese et al. ²⁴

Thirty‐two comorbidities (31 comorbidities included in the Elixhauser Index, and 1 unique comorbidity included in the Charlson Index) were determined via an in‐depth chart review analyzing anesthesia preoperative notes, medication lists, and encounter notes with other caretakers to determine status and severity.

Post‐CI speech recognition scores were collected clinically preoperatively and at 1, 3, 6, and/or 12 months postactivation, whereas PROMs were collected preoperatively and at 6 and/or 12‐month postactivation. All CI users were postlingually deafened adults with bilateral hearing loss qualifying for cochlear implantation in both ears, defined as unaided PTA > 60 dB HL and best‐aided CNC < 60% in both ears tested separately.

Speech Recognition Outcomes

Adult patients underwent audiologic evaluation according to the Minimum Speech Test Battery. ²⁵ Speech recognition testing for Consonant‐Nucleus‐Consonant words in quiet (CNC) and AzBio sentences in quiet (AzBioQ) was completed in a sound booth with speech presentation level of 60 dB sound pressure level (SPL, A‐weighted) ²⁶ , ²⁷ in the bilateral best‐aided condition. Stimuli were presented from a single loudspeaker at 0‐degree azimuth (1 m). Measurement for the best‐aided condition is consistent with the protocol described by McRackan et al. ²⁸

Patient‐Reported Outcome Measures

Self‐perceived hearing ability was evaluated using the Speech, Spatial and Qualities‐12 questionnaire (SSQ), ²⁹ a PROM that evaluates hearing across three domains: speech, spatial, and hearing qualities. The speech domain captures understanding in noisy or group settings. The spatial domain assesses sound localization, distance and movement perception, and spatial awareness. The hearing qualities domain addresses sound clarity, naturalness, ease of listening, and the ability to separate sounds. Each domain is scored from 0 (complete disability) to 10 (complete ability).

Hearing‐related QoL was assessed using the Cochlear Implant Quality of Life‐10 questionnaire (CIQOL). ³⁰ The CIQOL‐10 provides an overall evaluation of CI‐specific hearing‐related QoL. The questionnaire offers a single, comprehensive measure reflecting the broad impact of cochlear implantation on functional abilities and overall well‐being. Scores range from 0 (poorest QoL) to 50 (highest QoL).

Comorbidity Indices

The Charlson Comorbidity Index (Charlson Index) ¹⁹ is a widely used, weighted scoring system that predicts 10‐year mortality risk based on presence of chronic medical conditions. The index includes 16 comorbidities; each assigned a weight from 1 to 6 depending on its severity and association with long‐term mortality. For example, myocardial infarction, diabetes, and chronic pulmonary disease each carry a weight of 1, while conditions such as liver disease, solid tumors, and AIDS receive higher weights. The total score is calculated by summing the weights of all present conditions, ranging from 0 (no comorbidities) to 37. Higher scores reflect greater overall disease burden and a higher likelihood of mortality within 10 years. Notably, the index also considers age and assigns up to 4 points (for patients older than 80 years of age).

The Elixhauser Comorbidity Index (Elixhauser Index) ³¹ is a comprehensive, weighted tool designed to predict in‐hospital outcomes such as length of stay, cost, and mortality. It includes 31 comorbidities spanning a broad range of chronic conditions, including cardiac, pulmonary, endocrine, renal, psychiatric, and substance use disorders. Similarly to the Charlson Index, the Elixhauser index applies statistically derived weights to each comorbidity based on large‐scale hospital data. The weighted score implemented in this study reflects 30‐day hospital readmission. ²⁰ The total score can range from negative to positive values, with higher positive scores indicating greater predicted risk of adverse outcomes. Notably, some conditions (eg, hypothyroidism, uncomplicated hypertension) carry negative weights, reflecting potentially protective associations in certain populations.

The basic sum index used in this study assigned equal positive weights to the same 31‐comorbidiites included in the Elixhauser index and served as a reference for comparison with both the Charlson and Elixhauser indices.

Data Analysis

To aid in analysis and account for missing data, post‐CI speech recognition tasks collected at 1‐ or 3‐months post‐CI (whichever score was higher, as majority were collected at 3 months) were collapsed and treated as “early” outcomes, while postoperative speech recognition scores and PROM scores were collected at 6‐ or 12‐months post‐CI (whichever score was higher) were collapsed and treated as “long‐term” outcomes. This approach is supported given prior research suggesting a plateau in speech recognition outcomes at around 6 months of CI use. ³²

To observe general relationships between comorbidity indices and CI outcomes, bivariate analyses were conducted with Spearman‐rank order correlations, given that a few variables demonstrated nonnormal distributions based on visual inspection (eg, histograms, Q‐Q plots) and the Shapiro‐Wilk test. A false discovery rate (FDR) correction (⍺ = .05) was applied to account for multiple comparisons, ensuring control of the expected proportion of false positives while allowing for the identification of patterns in the data.

To isolate independent effects and provide potential clinical utility of each index, we ran additional exploratory analyses, conducting a series of multivariable regression analyses to predict each outcome with each index (18 total regressions; 6 unique outcomes and 3 index scores). Although some outcome variables demonstrate skewness (as stated above), residual diagnostics confirmed that model assumptions were sufficiently met, justifying the use of linear regression for multivariable modeling. We included key covariates known to influence CI outcomes and potential confounders of the relationship between comorbidity burden and outcomes. The key covariates included age, baseline pure‐tone average (PTA) in the ear to be implanted, baseline speech recognition score (CNC or AzBioQ) in best‐aided condition or baseline PROM score (SSQ or CIQOL‐10) for the predicted outcome, daily CI usage, and DoD.

Results

Summary of participant demographics and comorbidity index scores are shown in Table 1. None of the participants had prior CI surgery or history of CI revision. Pre‐CI, early post‐CI, and long‐term post‐CI outcomes are detailed in Table 2.

Table 1.

Mean (±SD) and Range of Demographic Characteristics and Comorbidity Sores

Characteristics	Mean (SD)	Range
Age (years)	66.19 (16.40)	17‐96
DoD (years)	5.92 (9.47)	0‐74
Long‐Term CI Usage (hours/day)	10.52 (3.60)	0.6‐18.2
PTA, ear to be implanted (dB)	87.59 (19.38)	28.33‐120
Charlson Index	4.08 (2.66)	0‐11
Elixhauser Index	−1.15 (4.65)	−14‐21
Basic Sum Index	2.30 (2.33)	0‐16

Table 2.

Mean (±SD) of CI Outcomes (CNC, AzBioQ, SSQ, and CIQOL‐10) at Pre‐CI, Early Post‐CI, and Long‐Term Post‐CI Time Points

	Pre‐CI	Early Post‐CI	Long‐term Post‐CI
CNC	34.4 (20.2)	58.1 (22.0)	67.1 (19.9)
AzBioQ	48.79 (28.00)	73.0 (23.5)	78.8 (22.7)
SSQ	1.9 (1.5)	—	5.0 (1.8)
CIQOL‐10	35.6 (8.1)	—	51.3 (10.4)

Bivariate Correlations

Results of bivariate analyses are shown in Table 3. The Charlson Index showed significant negative correlations after FDR correction with both early and long‐term post‐CI speech recognition outcomes, including CNC and AzBioQ scores (rs = –0.16 to –0.31). The Elixhauser Index was significantly correlated after FDR correction with early AzBioQ scores (rs = −0.22), but not with early CNC and long‐term outcomes. The basic sum index showed significant negative correlations after FDR correction with early CNC (rs = –0.15) and long‐term AzBioQ (rs = –0.21).

Table 3.

Spearman‐Rank Order Correlations Between Comorbidity Indices and Early and Long‐Term CI Outcomes, Including Speech Recognition Assessed in the Best‐Aided Condition and PROMs

			Charlson Index	Elixhauser Index	Basic Sum Index
Early Post‐CI	CNC	r	−0.16	−0.14	−0.15
		P‐value	0.008	0.026	0.012
	AzBioQ	r	−0.31	−0.22	−0.13
	—	P‐value	<0.001	0.002	0.064
Long‐Term Post‐CI	CNC	r	−0.24	−0.08	−0.09
	—	P‐value	<0.001	0.242	0.185
	AzBioQ	r	−0.30	−0.15	−0.21
	—	P‐value	<0.001	0.033	0.002
	SSQ	r	−0.04	−0.05	0.03
	—	P‐value	0.578	0.53	0.701
	CIQOL‐10	r	0.05	−0.07	0.06
	—	P‐value	0.654	0.475	0.535

Bolded values represent P‐values and correlations that remain significant after FDR correction (adjusting for 6 observations, P‐adjusted < .05).

None of the comorbidity indices demonstrated significant associations with PROMs, including the SSQ and CIQOL (Table 2).

Multivariable Regression

After adjusting for key clinical covariates (age, PTA, preoperative speech recognition (CNC or AzBioQ, depending on predicted outcome), daily CI usage, and DoD), only the Elixhauser Index emerged as a significant independent predictor of both early (β = –1.03, P = .005) and long‐term AzBioQ scores (β = –0.65, P = .050), but not CNC scores. The Charlson Index and basic sum index were not significant independent predictors of any outcomes in the multivariable models (Table 4).

Table 4.

Results From Series of Multivariate Regression Analyses (18 Total; 6 Unique CI Outcome Measures, 3 Unique Comorbidity Indices as Predictors)

			Charlson Index	Elixhauser Index	Basic Sum Index
Early Outcomes	CNC	β	−1.09	−0.59	−0.55
	—	P‐value	0.179	0.059	0.403
	AzBioQ	β	−1.61	−1.03	0.33
	—	P‐value	0.089	0.005	0.672
Long‐Term Outcomes	CNC	β	0.19	−0.12	0.25
	—	P‐value	0.795	0.696	0.687
	AzBioQ	β	−0.21	−0.65	−0.04
	—	P‐value	0.784	0.050	0.956
	SSQ	β	0.00	0.00	0.06
	—	P‐value	0.639	0.895	0.482
	CIQOL‐10	β	0.13	−0.02	0.10
	—	P‐value	0.883	0.951	0.880

Effect size and significance of comorbidity index terms are reported. Bolded values represent significant terms (P < .05).

None of the comorbidity indices were significant predictors of long‐term SSQ or CIQOL (Table 4).

Discussion

This study examined relationships between comorbidities and adult CI outcomes using weighted comorbidity indices. By applying the Charlson and Elixhauser Indices—tools previously validated in large‐scale medical populations—we aimed to clarify the role of overall disease burden in predicting auditory and patient‐reported outcomes following cochlear implantation. We first predicted that greater comorbidity burden, as reflected by a higher score on each index, would correlate with worse CI outcomes, including both speech recognition and PROM scores.

In line with our predictions, weighted comorbidity scores were significantly correlated with speech recognition outcomes in bivariate analyses, though no significant relationships with PROMs emerged. Specifically, the Charlson Index was consistently correlated with a medium effect size with both early and long‐term speech recognition outcomes. The stronger association of speech recognition with the Charlson Index, which predicts 10‐year mortality, suggests overlap between factors influencing long‐term survival and CI success.

However, when adjusting for clinical confounders, only the Elixhauser Index emerged as an independent predictor of speech recognition performance—specifically sentence recognition in quiet (AzBioQ) at both early and long‐term time points. This discrepancy highlights the value of index selection: while both scores reflect health status, they differ in scope and weighting. The Charlson Index assigns positive weights to a limited number of comorbidities and importantly considers age of participants when assigning a total index score. Thus, age‐related comorbidity and health status may not independently predict outcomes. To further evaluate this possibility, we conducted exploratory regressions excluding age from the models. When age was removed, the Charlson Index again reached statistical significance across speech recognition outcomes (Appendix A). This suggests that Charlson's apparent loss of significance in the primary models was largely due to collinearity with age, rather than absence of prognostic signal, although its predictive value is therefore not fully independent of age. Additionally, the inclusion of psychiatric and other chronic conditions not included in the Charlson Index (eg, hypothyroidism, rheumatoid arthritis) may better relate to CI outcomes.

Alternatively, the Elixhauser Index includes a broader set of comorbidities and assigns both positive and negative weights, potentially capturing more nuanced relationships between specific health conditions and CI outcomes. This unique feature may explain why the Elixhauser—but not Charlson—retained significance after covariate adjustment, and raises the potential for some comorbidities to have neutral or even beneficial effects on CI outcomes (evidenced by negative Elixhauser Index scores associated with higher AzBioQ scores), perhaps through increased social engagement or caregiver support.

These findings are consistent with previous literature. With regards to our findings with the Charlson Index, Aylward et al, found no relationship between an 11‐item frailty index and speech recognition outcomes post‐CI. ¹⁷ The frailty index, which in addition to assigning positive weights, only includes 11 comorbidities, supporting the need for more comprehensive indices that consider negative and positive associations with outcomes. Furthermore, prior work specifically comparing the Charlson and Elixhauser Indices in other surgical contexts has similarly found Elixhauser to outperform Charlson in predicting inpatient mortality and complications following orthopedic procedures. ³³ Overall, the superiority of the Elixhauser Index in our study and previous studies reflects the nuance and complexity in predicting chronic disease outcomes.

Interestingly, none of the comorbidity indices were significantly associated with PROMs, including the SSQ and CIQOL‐10. This finding diverges from Aylward et al, who reported a correlation between comorbidity burden and QoL using the Nijmegen Cochlear Implant Questionnaire. ¹⁷ Several factors may account for this discrepancy. First, the frailty index used in Aylward's study included only 11 items and uniformly applied positive weights, potentially overestimating the burden of certain conditions and lacking the nuance of more comprehensive indices. Second, our study employed the CIQOL‐10, a PROM specifically designed for CI users, which may more accurately capture functional improvements unique to this population. Third, we controlled for important covariates such as baseline QoL and CI usage in our multivariable models—an approach that reflects change in QoL rather than cross‐sectional associations.

Beyond these methodological differences, it is also possible that PROMs are less sensitive to physical or cognitive limitations that impact performance‐based outcomes like speech recognition, such as concentration and effort ablities. ³⁴ , ³⁵ PROMs instead may reflect psychosocial factors or expectations, which may be less directly tied to medical comorbidities. Further, The PROMs employed in this study, such as the SSQ, are specifically communication‐related and may not be as sensitive to physical health status. Though CIQOL‐10 was designed for hearing‐related QoL in CI users, it showed limited variability in our cohort compared to long‐term AzBioQ scores (SD, 6.92 vs 21.81). Future research should investigate these possibilities and explore whether specific domains of PROMs—such as listening‐effort or social engagement—are differentially affected by comorbidity burden in CI users.

To our knowledge, this is the first study to apply multiple weighted comorbidity indices to examine both speech recognition and PROMs following cochlear implantation. However, some limitations should be acknowledged. Our study is retrospective, and we were limited by missing data, resulting in post‐CI metrics being grouped into “early” and “long‐term” outcomes, though we would not expect large differences between the 1‐ and 3‐month outcomes, and 6‐ and 12‐month outcomes. Furthermore, there is potential inaccuracy when determining comorbidities from medical charts instead of collecting data from participants directly. Importantly, our preliminary study motivates the need to develop a weighted index specific to CI outcomes. Such a tool should incorporate conditions most relevant to auditory and communicative performance, beyond general medical morbidity. For example, chronic conditions that directly impact cognitive functioning (eg, dementia, cerebrovascular disease), neurologic processing (eg, traumatic brain injury), or sensory capacity (eg, visual impairment) may disproportionately affect CI rehabilitation and long‐term outcomes compared to other comorbidities. Similarly, psychiatric disorders (eg, depression, anxiety) and metabolic diseases (eg, diabetes, hypothyroidism) may indirectly influence outcomes through effects on neural plasticity, motivation, adherence to rehabilitation, or social engagement. Finally, sociodemographic and functional health factors (eg, daily device use, social support, frailty) may interact with medical comorbidities to shape real‐world performance. A CI‐specific index could therefore assign both positive and negative weights to such factors, reflecting not only disease burden but also potential protective effects (eg, caregiver support associated with certain chronic illnesses). Such refinements may improve prognostic accuracy, guide preoperative counseling, and inform individualized postoperative care plans.

Conclusion

This study represents the first efforts to examine the relative impact of different weighted comorbidity indices on CI outcomes. Findings highlight the influence of comorbid disease burden on speech recognition performance and suggest that certain comorbidities may even have protective or neutral effects. These preliminary results serve as the foundation for future work to develop a CI‐specific comorbidity index, tailored to the unique outcomes and mechanisms relevant to CI users. Such a tool could refine preoperative risk stratification, improve prognostic counseling, and guide post‐CI management in medically complex adults.

Author Contributions

Barak Spector, conceptualization, data collection, data analysis, interpretation of results, manuscript drafting, and revision; Priya Bandaru and Tam Nguyen, data extraction, literature review, and manuscript editing; Sandra Simmons, methodological guidance, interpretation of PROM‐related outcomes, and manuscript review; Aaron C. Moberly, clinical oversight, critical revision of the manuscript, and supervision of data interpretation; Terrin N. Tamat, study design, statistical oversight, data interpretation, and critical revision of the manuscript; all authors reviewed and approved the final manuscript.

Disclosures

Competing interests

A.C.M. is CMO and on the Board of Directors for Otologic Technologies, and a paid consultant for Amgen. The remaining authors have declared no potential conflicts of interest.

Funding source

T.N.T. has received research funding through Cochlear Americas.

Supporting information

Appendix A: Results from series of multivariate regression analyses predicting CI outcomes measures with the Charlson Index excluding age as a key covariate. Effect size and significance of comorbidity index terms are reported. Bolded values represent significant terms (p < .05).