청력과 음성 지표의 통합분석을 통한 치매 선별 가능성 : Integrated Analysis of Hearing and Voice Measures for Dementia Screening

다국어 초록 (Multilingual Abstract)

Dementia is a progressive neurocognitive disorder characterized by declines in multiple cognitive domains, including memory, language, and executive function. Early identification of individuals at risk for dementia is essential for timely diagnosis and intervention. However, commonly used cognitive screening tests, comprehensive neuropsychological assessments, and neuroimaging-based evaluations have limited applicability as large-scale screening tools because of their dependence on trained professionals, time requirements, cost, and limited accessibility.
Accordingly, increasing attention has been directed toward non-invasive functional measures that can be repeatedly obtained and used to identify older adults who may require further cognitive and medical evaluation. Hearing loss has been associated with cognitive decline and an increased risk of dementia, whereas acoustic voice measures may reflect changes in phonatory stability and cognitive-motor control. However, hearing and voice measures have generally been investigated independently, and the potential value of integrating these measures for dementia screening remains unclear.
The purpose of this study was to examine hearing and voice characteristics in older adults with dementia and to investigate their potential utility as complementary screening measures. Hearing thresholds and acoustic voice measures were compared between 20 individuals with dementia and 20 cognitively normal older adults. The relationships between hearing and voice measures were also examined. In addition, the extent to which hearing and voice measures explained cognitive performance and discriminated between individuals with dementia and cognitively normal older adults was evaluated.
Hearing was assessed using the three-frequency pure-tone average （PTA3） and four-frequency pure-tone average （PTA4）. Voice was assessed during sustained vowel phonation and connected speech using acoustic voice measures. The sustained vowel measures included fundamental frequency, Jitter, Shimmer, noise-to-harmonic ratio, cepstral peak prominence （CPP）, low-to-high spectral ratio, and cepstral spectral index of dysphonia （CSID）. For the connected speech task, CPP, low-to-high spectral ratio, and CSID were also analyzed. Cognitive performance was assessed using the Mini-Mental State Examination for Dementia Screening （MMSE-DS）.
The results showed that the dementia group had significantly higher PTA3 and PTA4 values than the cognitively normal group, indicating poorer hearing sensitivity across the speech-frequency range. These findings suggest that elevated pure-tone thresholds are a characteristic feature of the older adults with dementia included in this study.
Regarding acoustic voice measures, the dementia group demonstrated significantly higher Jitter and CSID values and significantly lower CPP values during sustained vowel phonation. These findings indicate increased cycle-to-cycle irregularity, reduced cepstral prominence, and poorer overall voice quality in the dementia group. No significant group differences were found for fundamental frequency, Shimmer, noise-to-harmonic ratio, or low-to-high spectral ratio.
In contrast, group differences in CPP, low-to-high spectral ratio, and CSID during connected speech were not significant. The finding that significant differences were observed during sustained vowel phonation, but not during connected speech, suggests that acoustic manifestations of dementia may vary according to the speech task. Sustained vowel phonation may be more sensitive to subtle phonatory instability because it permits the assessment of vocal-fold vibration under relatively controlled conditions.
Significant relationships were also observed between hearing thresholds and sustained vowel measures. Higher PTA3 and PTA4 values were associated with increased Jitter and CSID and with decreased CPP. These results suggest that poorer hearing sensitivity was related to greater acoustic instability during sustained phonation. No consistent relationships were observed between the hearing measures and the acoustic parameters obtained from connected speech.
PTA3, PTA4, and sustained vowel CSID each significantly explained MMSE-DS scores when examined as individual predictors. Higher hearing thresholds and greater CSID values were associated with lower cognitive scores. The hearing measures continued to provide significant explanatory information after years of education were considered. Sustained vowel CSID also contributed significantly when it was entered with education alone.
When the hearing and voice measures were entered together, PTA3 and PTA4 remained significant predictors of MMSE-DS performance, whereas the independent contribution of CSID was not significant. Adding CSID to the hearing-based models did not produce a statistically significant increase in explained variance. These findings indicate that hearing thresholds accounted for a larger proportion of the information shared with cognitive performance, while CSID provided limited additional explanatory value after hearing was considered.
Receiver operating characteristic analyses showed that PTA3, PTA4, Jitter, CPP, and sustained vowel CSID significantly distinguished the dementia group from the cognitively normal group. Among the individual measures, PTA3 and PTA4 demonstrated the strongest screening performance, with area under the curve values of .879 and .893, respectively. Sustained vowel CSID showed the highest screening performance among the voice measures, with an area under the curve of .800.
The education-adjusted models combining hearing and voice measures yielded high overall screening performance. The model including education, PTA3, and sustained vowel CSID produced an area under the curve of .933, whereas the model including education, PTA4, and sustained vowel CSID produced an area under the curve of .943. However, the addition of CSID did not significantly improve the area under the curve compared with the corresponding hearing-based models, regardless of whether education was included.
Taken together, these findings indicate that older adults with dementia exhibit both elevated pure-tone hearing thresholds and reduced acoustic stability during sustained vowel phonation. Hearing loss and phonatory instability were significantly related, and both functional domains were associated with cognitive performance. However, pure-tone averages demonstrated more consistent explanatory and screening performance than the acoustic voice measures.
The limited additional contribution of CSID in the integrated models suggests that combining hearing and voice measures does not necessarily produce a statistically significant improvement over hearing-based screening alone. Voice measures may nevertheless provide complementary information about phonatory control that is not directly represented by pure-tone thresholds. In particular, Jitter, CPP, and CSID obtained from sustained vowel phonation appear to be more sensitive to dementia-related acoustic differences than the corresponding measures obtained from connected speech.
Overall, pure-tone hearing measures showed promise as accessible functional indicators for identifying older adults who may require further cognitive assessment. Acoustic voice measures derived from sustained vowel phonation may provide complementary indicators of phonatory instability associated with dementia. Although the integrated assessment of hearing and voice did not significantly improve screening performance beyond hearing measures alone, it may offer additional functional information and serve as a non-invasive adjunctive approach for the preliminary screening of dementia. Such measures should be considered complementary to, rather than replacements for, comprehensive cognitive and medical evaluations.

목차 (Table of Contents)

• Ⅰ. 서론 1
• 1. 연구의 배경 및 목적 1
• 2. 연구 문제 8
• Ⅱ. 이론적 배경 10
• 1. 치매 10
• 1.1. 치매의 정의 10
• 1.2. 치매의 역학적 특성 10
• 1.3. 치매의 주요 유형 11
• 1.4. 치매의 진행 단계 및 증상 12
• 2. 치매의 진단 13
• 2.1. 치매의 임상적 진단 기준 13
• 2.2. 치매 진단을 위한 인지기능 평가 15
• 가. 간이정신상태검사 15
• 나. 서울신경심리검사 16
• 2.3. 치매의 중증도 평가 17
• 3. 노인성 난청과 청력 지표 19
• 3.1. 노인성 난청의 정의 및 특성 19
• 3.2. 순음청력검사와 평균순음청력역치 19
• 3.3. 평균순음청력역치 산출 방법 21
• 가. 3분법 21
• 나. 4분법 21
• 다. 6분법 22
• 3.4. 청력 저하와 인지기능의 관련성 23
• 4. 치매와 음성 및 음질 지표 24
• 4.1. 치매의 말·언어 및 음성 특성 24
• 4.2. 음질의 정의와 음향학적 평가 26
• 4.3. 음질 지표 27
• 가. 성대 진동 관련 지표 27
• 나. 켑스트럼 및 스펙트럼 지표 29
• 4.4. 발화 과제에 따른 음질 특성 32
• 가. 모음 연장발성 32
• 나. 연결발화 33
• 4.5. 치매 집단의 음향학적 음성 특성 34
• Ⅲ. 연구 방법 37
• 1. 연구 참여자 37
• 1.1. 치매 노인 집단 37
• 1.2. 정상 노인 집단 38
• 1.3. 연구 참여자의 일반적 특성 40
• 2. 연구 절차 및 연구 과제 41
• 2.1. 연구 절차 41
• 2.2. 연구 과제 42
• 가. 청력역치 과제 42
• 나. 음질 지표 과제 42
• 3. 연구 도구 43
• 3.1. 순음청력검사기 43
• 3.2. 다차원적 음성 프로그램 43
• 3.3. 켑스트럼 기반 음성 프로그램 44
• 4. 자료 분석 44
• 4.1. 순음청력검사 44
• 4.2. 음질 지표 분석 45
• 5. 통계 분석 46
• 5.1. 연구 참여자의 일반적 특성 및 집단 비교 46
• 5.2. 청력 지표와 음성 지표 간 상관분석 47
• 5.3. MMSE-DS 설명을 위한 선형회귀분석 47
• 5.4. 치매 선별을 위한 ROC 및 로지스틱 회귀분석 48
• 5.5. 단일모형과 통합모형 간 AUC 비교 48
• 5.6. 신뢰도 분석 49
• 가. 검사자 내 신뢰도 49
• 나. 검사자 간 신뢰도 50
• Ⅳ. 연구 결과 52
• 1. 분석 전 정규성 검정 결과 52
• 2. 청력 및 음성 특성 비교 결과 52
• 2.1. 집단 간 청력역치 결과 52
• 2.2. 집단 간 모음 연장발성 음성 측정치 결과 54
• 2.3. 집단 간 연결발화 음성 측정치 결과 59
• 3. 청력 지표와 음성 지표 간 상관관계 60
• 4. 청력 및 음성 지표의 MMSE-DS 설명력 61
• 4.1. 교육연수를 보정하지 않은 선형회귀분석 결과 61
• 4.2. 교육연수를 보정한 위계적 선형회귀분석 결과 64
• 5. 청력 및 음성 지표의 치매 선별 정확도 66
• 5.1. 단일 지표의 ROC 분석 결과 66
• 5.2. 교육연수를 보정하지 않은 통합 로지스틱 회귀분석 결과 68
• 5.3. 교육연수를 보정한 로지스틱 회귀분석 결과 71
• 5.4. 단일모형과 통합모형 간 AUC 차이 검정 결과 75
• Ⅴ. 논의 77
• 1. 치매 노인의 청력 및 음성 특성 77
• 1.1. 치매 노인의 청력 저하 특성 77
• 1.2. 모음 연장발성에서 나타난 음향학적 불안정성 79
• 1.3. 발화 과제에 따른 음성 특성 81
• 2. 청력과 음성 특성의 관련성 82
• 3. 청력 및 음성 지표와 인지기능의 관계 84
• 4. 치매 선별 지표로서의 활용 가능성 85
• 4.1. 단일 청력 및 음성 지표의 선별 성능 85
• 4.2. 청력과 음성 지표의 통합 선별 성능 86
• 5. 연구의 제한점 및 제언 88
• 6. 결론 89
• 참고문헌 91
• 감사의 글 99