The association between Naples prognostic score and cognitive function in older adults

Abstract

The purpose of this study was to examine the association between the Naples prognostic score (NPS) and cognitive function in older adults. This cross-sectional study analyzed data from 2759 participants aged 60 years and older, with a mean age of 69.2 ± 6.6 years, from the National Health and Nutrition Examination Survey 2011 to 2014 data. Participants were categorized into 2 groups based on NPS: Group 1 (NPS = 0–2) and Group 2 (NPS = 3–4). Cognitive function was assessed using the Consortium to Establish a Registry for Alzheimer’s Disease Word Learning subtest, Animal Fluency Test (AFT), and Digit Symbol Substitution Test (DSST). A global composite z-score was also calculated. Multivariable linear and logistic regression models were used to examine the associations between NPS and cognitive outcomes, with adjustments for sociodemographic and health-related confounders. In the fully adjusted models, a higher NPS was associated with lower AFT, DSST, and composite z-scores, whereas no significant association was observed with the Consortium to Establish a Registry for Alzheimer’s Disease Word Learning subtest score. Compared with the low-NPS group, the high-NPS group had significantly lower AFT scores (β = −1.39, 95% confidence interval [CI]: −1.89 to −0.88), DSST scores (β = −1.82, 95% CI: −3.11 to −0.53), and composite z-scores (β = −0.43, 95% CI: −0.62 to −0.24), all with P < .05. Logistic regression further showed that the high-NPS group had higher odds of cognitive impairment according to the AFT (odds ratio = 1.41, 95% CI: 1.04–1.92) and DSST (odds ratio = 1.47, 95% CI: 1.00–2.14), both with P < .05. No statistically significant interactions were detected in the subgroup analyses (all P for interaction > .05). This study revealed the possible association between a high NPS and poorer cognitive performance, especially in terms of verbal fluency and processing speed. These findings suggest that the NPS may provide supplementary information for identifying individuals with poorer cognitive performance. However, the cross-sectional design precludes temporal or causal inference.

1. Introduction

Cognitive impairment is common among older adults and represents an important public health concern. In a nationally representative US study, Manly et al reported that among adults aged 65 years and older, the estimated prevalence of mild cognitive impairment was 22%, while the prevalence of dementia was 10%, indicating that approximately one-third had either condition.^[1] Cognitive impairment may affect multiple domains, including memory, attention, language, executive function, processing speed, reasoning, and visuospatial ability, and may adversely affect daily functioning and quality of life. Mild cognitive impairment is generally regarded as an intermediate clinical state between normal cognitive aging and dementia and is associated with Alzheimer disease (AD) pathology and cerebral infarction.^[2] Consequently, the identification of accessible biomarkers associated with early cognitive impairment may facilitate the timely identification and further evaluation of older adults at risk of cognitive decline.^[3]

Systemic inflammation has the potential to induce a proinflammatory milieu within the central nervous system, with both inflammation and the immune system being increasingly acknowledged as critical contributors to cognitive decline and the progression of dementia.^[4,5] Recent research has indicated that various systemic inflammation-related markers, including the systemic inflammation response index, systemic immune–inflammation index, and pan-immune–inflammation value, may be correlated with age-related cognitive decline.^[6,7] Furthermore, certain nutritional markers, including decreased serum albumin (ALB) levels and elevated total cholesterol (TC) levels, have been linked to cognitive impairment in the older population.^[8,9] Other inflammation-related indicators, including the neutrophil-to-lymphocyte ratio (NLR) and lymphocyte-to-monocyte ratio (LMR), have also been associated with cognitive function.^[10,11] However, most previous studies have examined individual inflammatory or nutritional biomarkers rather than a composite indicator integrating both dimensions. The Naples prognostic score (NPS), introduced by Galizia et al in 2017, combines ALB, TC, the NLR, and the LMR and has been widely investigated as a prognostic indicator in cancer and other chronic diseases.^[12–15] A recent multicenter study by Ran et al reported an association between a higher NPS and cognitive impairment in patients undergoing hemodialysis.^[16] Nevertheless, whether the NPS is associated with cognitive function in the general older population and across different cognitive domains remains unclear. Therefore, this study used nationally representative National Health and Nutrition Examination Survey (NHANES) 2011 to 2014 data to examine the associations of the NPS with verbal learning and memory, verbal fluency, processing speed, and global cognitive performance in US adults aged 60 years and older. This approach extends previous research by evaluating the NPS in a general older population rather than in a specific clinical group.

2. Materials and methods

2.1. Study population

This study was reported in accordance with the Strengthening the Reporting of Observational Studies in Epidemiology guidelines.^[17] The NHANES data were accessed and analyzed on December 16, 2024. As shown in Figure 1, this study focused on NHANES 2011 to 2014 participants aged 60 years and older (n = 3632). The final analysis included 2759 eligible participants after participants with missing cognitive function data (n = 698) and NPS assessment data (ALB, TC, and complete blood count; n = 175) were removed. All NHANES participants provided written informed consent. The NHANES 2011 to 2014 data were collected under protocols approved by the National Center for Health Statistics Research Ethics Review Board, Centers for Disease Control and Prevention. The NHANES 2011 to 2012 cycle was approved under Protocol No. 2011-17, and the NHANES 2013 to 2014 cycle was conducted as a continuation of Protocol No. 2011-17. Because this study used publicly available, deidentified NHANES data, additional institutional ethics approval was not required.

Figure 1.

A flowchart showing the selection of the participants. LMR = lymphocyte-to-monocyte ratio, NHANES = National Health and Nutrition Examination Survey, NLR = neutrophil-to-lymphocyte ratio, TC = total cholesterol.

2.2. Assessment of the NPS

Galizia et al^[12] developed the NPS, which is based on ALB, TC, the NLR, and the LMR. According to previous studies,^[13] an ALB level ≥ 40 g/L, TC > 180 mg/dL, NLR < 2.96, or LMR > 4.44 was assigned a score of 0, whereas an ALB level < 40 g/L, TC ≤ 180 mg/dL, NLR ≥ 2.96, or LMR ≤ 4.44 was assigned a score of 1. The NPS was calculated by summing the scores of these 4 parameters. Furthermore, participants were categorized into 2 groups based on the NPS, as outlined in a previous study: patients with scores of 0, 1, or 2 were classified as having a low NPS, while those with NPS values of 3 or 4 were categorized as having a high NPS.

2.3. Cognitive function assessment

A series of cognitive tests were administered to NHANES (2011–2014) participants aged >60 years. These assessments included the Consortium to Establish a Registry for Alzheimer’s Disease (CERAD) test, the Animal Fluency Test (AFT), and the Digit Symbol Substitution Test (DSST). The CERAD Word Learning subtest (CERAD W-L) evaluates both immediate and delayed learning capabilities for novel verbal information, which is a critical aspect of the memory subdomain.^[18] This subtest comprises 3 consecutive learning trials, referred to as the Immediate Word Recall Test, followed by a delayed recall trial known as the Delayed Word Recall Test. During the learning trials, the participants were instructed to read aloud a set of 10 unrelated words and subsequently recall as many of these words as possible immediately. The sequence of the 10 words varied across each learning trial. Delayed word recall was conducted following the completion of 2 additional cognitive assessments: the AFT and the DSST. The total CERAD W-L score was calculated by summing the scores from the 3 immediate-recall trials and the delayed-recall trial, yielding a possible score range of 0 to 40. Higher scores indicated better verbal learning and memory.^[19]

Categorical verbal fluency, an aspect of executive function, was assessed utilizing the AFT. The participants were allotted 1 minute to name as many animals as possible, with each correctly named animal receiving 1 point. The AFT has no fixed maximum score, and higher scores indicate better categorical verbal fluency and executive function.^[19] In NHANES, participants were initially prompted to name 3 items of clothing as a preparatory exercise prior to undertaking the AFT.

The DSST assesses executive function, visual-spatial ability, and information processing speed. In this test, participants are provided with a sheet of paper displaying a key at the top, which pairs 9 numbers with corresponding symbols. The participants are required to draw the symbols corresponding to each number within a 2-minute time frame. The score ranges from 0 to 133 and is determined by the total number of correct symbol–number matches, with higher scores indicating better processing speed, attention, and executive function.^[19]

To eliminate dimensional differences in tests and quantify impairments in cognition, in this study, the sum of the 3 standardized z-scores (CERAD W-L, AFT, and DSST) is recorded as an index for global cognition.^[20] The formula for calculating a z-score is z = (x − μ)/σ, where x is the raw score, μ is the population mean, and σ is the population standard deviation. Higher composite z-scores indicated better overall cognitive performance, whereas lower scores indicated poorer cognitive performance. Scores derived from the cognitive tests were regarded as continuous variables and were also dichotomized, with the lowest nonsurvey-weighted quartile defining cognitive function impairment.

2.4. Covariates

The potential covariates were selected on the basis of existing knowledge and literature on cognitive function. The covariates considered in this study included age (categorized as 60–69, 70–79, and ≥80 years), sex (male and female), race (non-Hispanic White, non-Hispanic Black, Mexican American, and Other races), education level (less than high school, high school or equivalent, and college or above), marital status (married, previously married, and never married), the household poverty-income ratio (PIR; ≤1.30, 1.31–3.5, >3.5, and others), body mass index (BMI; <25 kg/m², 25–30 kg/m², and ≥30 kg/m²), smoking status (yes or no), alcohol consumption status (yes or no), engagement in moderate recreational activities (yes or no), and medical history of hypertension, diabetes, heart disease, and stroke.

Individuals classified as never smokers were those who reported a lifetime consumption of fewer than 100 cigarettes. Current smokers were identified as individuals who had smoked more than 100 cigarettes throughout their lifetime, whereas former smokers were those who had also smoked more than 100 cigarettes but subsequently ceased smoking.^[21] Excessive alcohol consumption was characterized as an intake of 10 grams or more per day (equivalent to at least 1 drink) for women and 20 grams or more per day (equivalent to at least 2 drinks) for men.

2.5. Statistical analysis

In accordance with the intricate sampling design and data clustering, all analyses were conducted using weighted methods in R software (http://www.R-project.org; The R Foundation, Austria), STATA 17.0 (Stata Corp, College Station), and Empower Stats (http://www.empowerstats.com; X&Y Solutions, Inc.). Owing to the integration of 2 cycles of NHANES data, the original sample weight, WTMEC2YR, was adjusted by dividing it by 2.^[22] Baseline characteristics for categorical variables are given as weighted percentages (95% confidence interval [CI]) and were compared through the χ² test. Continuous variables are presented as the mean ± standard deviation and were analyzed through weighted linear regression. Multivariate logistic and linear regression analyses were applied to study the relationships between the NPS and cognitive function. The distributions of continuous variables were assessed using Q–Q plots and formal normality testing. Because the individual NPS components and most cognitive test scores showed nonnormal distributions, Spearman rank correlation was used to evaluate correlations between the individual NPS components and cognitive measures. Variance inflation factors were calculated to assess multicollinearity. Three models were constructed: Model 1 was unadjusted; Model 2 was adjusted for age, sex, and race; and Model 3 was additionally adjusted for education level, PIR, BMI, alcohol consumption, smoking status, physical activity, hypertension, heart disease, stroke, and diabetes. Finally, to evaluate the consistency of the association between the NPS and cognitive function, we undertook further subgroup analyses and interaction tests, adjusting for every confounding factor. Missing data were handled by coding them as a missing indicator category for categorical variables and using mean values for continuous variables. A P value of <.05 was considered to indicate statistical significance.

3. Results

3.1. Characteristics of the study participants

On the basis of the exclusion criteria in Figure 1, we selected 2759 older participants from NHANES 2011 to 2014, with a mean age of 69.2 ± 6.6 years (53.8% females, 46.2% males). The characteristics of all the participants, of whom 2258 (81.8%) were considered to be in the low-NPS group and 501 (18.2%) were in the high-NPS group, are summarized in Table 1. Compared with those in the low-NPS group, the high-NPS group participants were typically older males with lower education level, income, and a higher prevalence of smoking, physical inactivity, hypertension, diabetes, heart disease, and stroke. Additionally, people were more prone to having lower TC, ALB, and LMR levels and higher NLR levels. Moreover, the high-NPS group had lower scores on all 3 cognitive function tests and lower composite z-scores than did the low-NPS group.

Table 1.

Characteristics of included participants in NHANES 2011 to 2014.

Characteristics	Total (N = 2759)	Low NPS (N = 2258)	High NPS (N = 501)	P value
Gender				<.001
Male	46.2 (43.6–48.8)	43.4 (40.5–46.2)	60.0 (54.4–65.4)
Female	53.8 (51.2–56.4)	56.6 (53.8–59.5)	40.0 (34.6–45.6)
Age (yr)	69.2 ± 6.6	68.7 ± 6.5	71.8 ± 6.8	<.001
60–69	56.1 (53.6–58.6)	59.6 (56.9–62.3)	39.0 (33.2–45.1)
70–79	29.3 (27.2–31.6)	28.5 (26.1–30.9)	33.6 (28.5–39.1)
≥80	14.5 (13.1–16.1)	11.9 (10.5–13.5)	27.4 (23.1–32.2)
Race/ethnicity				.982
Non-Hispanic White	80.3 (78.9–81.7)	80.2 (78.7–81.7)	80.6 (77–83.8)
Non-Hispanic Black	7.9 (7.2–8.6)	7.9 (7.1–8.7)	7.8 (6.2–9.7)
Mexican American	3.4 (2.9–3.8)	3.3 (2.9–3.9)	3.5 (2.6–4.8)
Other races*	8.5 (7.5–9.5)	8.5 (7.5–9.6)	8.0 (5.9–10.9)
Marital status				.915
Married/cohabitation	65.0 (62.6–67.3)	65.2 (62.6–67.8)	63.7 (58.3–68.9)
Was married	30.6 (28.4–32.9)	30.4 (28.0–32.9)	31.8 (26.9–37.1)
Never married	4.4 (3.6–5.4)	4.4 (3.5–5.5)	4.5 (2.7–7.3)
Education level				.003
Less than high school	15.7 (14.2–17.3)	15.1 (13.5–16.9)	18.6 (15.0–22.8)
High school	22.3 (20.3–24.5)	21.4 (19.2–23.8)	26.9 (22.1–32.4)
College or higher	61.9 (59.5–64.3)	63.5 (60.8–66.1)	54.4 (48.7–60.0)
Poverty-income ratio	3.1 ± 1.6	3.2 ± 1.6	2.8 ± 1.5	<.001
<1.3	16.4 (14.9–18.0)	15.8 (14.3–17.5)	19.1 (15.5–23.2)
1.3–3.5	35.7 (33.3–38.2)	34.2 (31.5–36.9)	43.4 (37.8–49.1)
≥3.5	41.6 (39–44.2)	43.4 (40.6–46.4)	32.7 (27.1–38.7)
Not recorded	6.3 (5.2–7.5)	6.6 (5.4–8.0)	4.9 (3.0–8.0)
BMI (kg/m2)				.437
<25 kg/m2	26.0 (23.8–28.3)	25.9 (23.5–28.4)	26.4 (21.7–31.7)
25–30 kg/m2	37.1 (34.7–39.6)	37.6 (34.9–40.4)	34.6 (29.3–40.3)
≥30 kg/m2	36.9 (34.5–39.4)	36.5 (33.8–39.3)	39.0 (33.5–44.8)
Physical activity level				<.001
Inactive	38.3 (35.9–40.8)	37.0 (34.3–39.7)	44.8 (39.2–50.5)
Less active	16.2 (14.4–18.1)	15.8 (13.9–17.9)	18.3 (14.2–23.2)
Active	45.4 (42.8–48)	47.2 (44.3–50.1)	36.6 (31.1–42.4)
Drinking status				.333
Yes	72.0 (69.7–74.1)	72.4 (69.9–74.7)	70.2 (64.8–75)
No	26.7 (24.6–28.9)	26.2 (23.9–28.7)	29.0 (24.2–34.3)
Smoking status				<.001
Never smoker	49.8 (47.3–52.4)	51.8 (48.9–54.6)	40.4 (34.9–46.1)
Former smoker	38.9 (36.5–41.5)	37.0 (34.3–39.8)	48.1 (42.4–53.9)
Current smoker	11.2 (9.7–12.9)	11.2 (9.5–13.1)	11.4 (8.2–15.8)
Hypertension, n (%)				<.001
Yes	58.3 (55.7–60.9)	56.5 (53.6–59.3)	67.5 (61.6–72.8)
No	41.4 (38.9–44)	43.2 (40.4–46.1)	32.5 (27.2–38.4)
Diabetes, n (%)				<.001
Yes	23.0 (21.0–25.1)	20.7 (18.6–23.0)	34.3 (29.2–39.7)
No	77.0 (74.9–79.0)	79.3 (77.0–81.4)	65.7 (60.2–70.7)
Heart disease, n (%)				<.001
Yes	9.7 (8.3–11.2)	7.7 (6.4–9.3)	19.1 (14.9–24.2)
No	90.3 (88.7–91.7)	92.2 (90.7–93.6)	80.9 (75.8–85.1)
Stroke, n (%)				<.001
Yes	6.5 (5.4–7.8)	5.2 (4.1–6.6)	13.0 (9.7–17.1)
No	93.4 (92.0–94.5)	94.7 (93.3–95.8)	87.0 (82.9–90.3)
TC (mg/dL)	193.0 ± 42.4	196.9 ± 41.1	157.1 ± 32.4	<.001
ALB (g/L)	42.1 ± 2.9	42.4 ± 2.6	40.0 ± 3.4	<.001
NLR	2.5 ± 1.6	2.3 ± 1.2	4.0 ± 2.3	<.001
LMR	3.6 ± 1.7	3.8 ± 1.7	2.5 ± 0.9	<.001
CERAD W-L score	32.2 ± 8.5	32.7 ± 8.5	29.7 ± 8.3	<.001
AFT score	18.1 ± 5.6	18.5 ± 5.6	15.9 ± 4.9	<.001
DSST score	52.2 ± 16.7	53.3 ± 16.6	45.5 ± 15.4	<.001
Composite z-score	0.8 ± 2.4	1.0 ± 2.4	−0.3 ± 2.2	<.001

Values are weighted mean ± standard deviation or weighted % (95% confidence interval). P values are weighted.

The composite z-score was calculated by summing the z-scores ([test score − mean score]/SD) of the 3 individual tests.

AFT = Animal Fluency Test, ALB = serum albumin, BMI = body mass index, CERAD W-L = Consortium to Establish a Registry for Alzheimer’s Disease Word Learning subtest, DSST = Digit Symbol Substitution Test, LMR = lymphocyte-to-monocyte ratio, NHANES = National Health and Nutrition Examination Survey, NLR = neutrophil-to-lymphocyte ratio, NPS = Naples prognostic score, SD = standard deviation, TC = total cholesterol.

^*Other races include American Indian or Alaska Native, Native Hawaiian or Other Pacific Islander, and multiracial persons.

3.2. Correlations between individual NPS components and cognitive function

Spearman correlations between the 4 individual NPS components (NLR, LMR, TC, and ALB) and the 4 cognitive measures (CERAD W-L, AFT, DSST, and the composite z-score) are presented in Table 2. NLR was negatively correlated with CERAD W-L (ρ = −0.109, P < .001) and the composite z-score (ρ = −0.050, P = .009), whereas its correlations with AFT and DSST were not statistically significant. LMR was positively correlated with CERAD W-L (ρ = 0.108, P < .001) and the composite z-score (ρ = 0.049, P = .011), but not with AFT or DSST. TC was positively correlated with all 4 cognitive measures (all P < .001). ALB was also positively correlated with CERAD W-L (ρ = 0.056, P = .0035), AFT (ρ = 0.095, P < .001), DSST (ρ = 0.108, P < .001), and the composite z-score (ρ = 0.107, P < .001).

Table 2.

Correlation between individual indicators in NPS and cognition.

Individual indicator	Cognitive function	Spearman ρ	P value
NLR	CERAD W-L	−0.109	<.001
	AFT	−0.002	.936
	DSST	−0.017	.367
	Composite z-score	−0.050	.009
LMR	CERAD W-L	0.108	<.001
	AFT	−0.005	.797
	DSST	0.023	.233
	Composite z-score	0.049	.011
TC	CERAD W-L	0.123	<.001
	AFT	0.101	<.001
	DSST	0.126	<.001
	Composite z-score	0.141	<.001
ALB	CERAD W-L	0.056	.0035
	AFT	0.095	<.001
	DSST	0.108	<.001
	Composite z-score	0.107	<.001

The symbol ρ denotes the Spearman rank correlation coefficient. The composite z-score was calculated by summing the z-scores ([test score − mean score]/SD) of the 3 individual tests.

AFT = Animal Fluency Test, ALB = serum albumin, CERAD W-L = Consortium to Establish a Registry for Alzheimer’s Disease Word Learning subtest, DSST = Digit Symbol Substitution Test, LMR = lymphocyte-to-monocyte ratio, NLR = neutrophil-to-lymphocyte ratio, NPS = Naples prognostic score, SD = standard deviation, TC = total cholesterol.

3.3. Associations between the NPS and cognitive impairment

All the factors showed variance inflation factor < 2. After that, we used a multivariate linear regression model to examine the relationships between the NPS and cognitive tests as continuous variables (Table 3). The explanatory power of the fully adjusted models varied across the cognitive outcomes, with adjusted R² values of 0.222 for the CERAD W-L score, 0.275 for the AFT score, 0.466 for the DSST score, and 0.438 for the composite z-score when NPS was analyzed as a continuous variable. After adjusting for relevant covariates (age, sex, race, the PIR, physical activity status, BMI, education level, smoking status, alcohol consumption status, diabetes status, hypertension status, stroke status, and heart disease status), we detected associations between a higher NPS and lower cognitive performance measured by the AFT (β = −0.58, 95% CI: −0.78 to −0.37, P < .001), DSST (β = −0.85, 95% CI: −1.38 to −0.33, P = .002), and composite z-scores (β = −0.18, 95% CI: −0.26 to −0.11, P < .001). In Model 3, no significant association was found between the NPS and cognitive performance measured using the CERAD W-L (Table S1, Supplemental Digital Content 1). Additionally, the results revealed similar outcomes when the NPS was divided into 2 levels. Compared with the low-NPS group, the high-NPS group presented significantly lower AFT scores (β = −1.39, 95% CI: −1.89 to −0.88, P < .001) and DSST scores (β = −1.82, 95% CI: −3.11 to −0.53, P = .006). Furthermore, a higher NPS was also associated with a lower global cognitive z-score (β = −0.43, 95% CI: −0.62 to −0.24, P < .001).

Table 3.

Linear regression analysis for associations between NPS and cognitive function.

	Model 1	Model 2	Model 3
	β (95% CI), P	β (95% CI), P	β (95% CI), P
CERAD W-L score
Continuous
NPS	−1.44 (−1.76 to −1.12), <.001	−0.49 (−0.81 to −0.18), .002	−0.26 (−0.58 to 0.06), .114
Model fit	Adjusted R2 = 0.018	Adjusted R2 = 0.167	Adjusted R2 = 0.222
Categorical
Low NPS	Ref	Ref	Ref
High NPS	−3.01 (−3.84 to −2.17), <.001	−1.03 (−1.82 to −0.23), .011	−0.60 (−1.39 to 0.19), .134
Model fit	Adjusted R2 = 0.003	Adjusted R2 = 0.166	Adjusted R2 = 0.221
AFT score
Continuous
NPS	−1.00 (−1.22 to −0.79), <.001	−0.80 (−1.01 to −0.59), <.001	−0.58 (−0.78 to −0.37), <.001
Model fit	Adjusted R2 = 0.015	Adjusted R2 = 0.169	Adjusted R2 = 0.275
Categorical
Low NPS	Ref	Ref	Ref
High NPS	−2.61 (−3.16 to −2.06), <.001	−1.87 (−2.40 to −1.35), <.001	−1.39 (−1.89 to −0.88), <.001
Model fit	Adjusted R2 = 0.003	Adjusted R2 = 0.161	Adjusted R2 = 0.273
DSST score
Continuous
NPS	−3.60 (−4.23 to −2.96), <.001	−1.94 (−2.52 to −1.36), <.001	−0.85 (−1.38 to −0.33), .002
Model fit	Adjusted R2 = 0.171	Adjusted R2 = 0.282	Adjusted R2 = 0.466
Categorical
Low NPS	Ref	Ref	Ref
High NPS	−8.07 (−9.71 to −6.44), <.001	−4.06 (−5.51 to −2.61), <.001	−1.82 (−3.11 to −0.53), .006
Model fit	Adjusted R2 = 0.001	Adjusted R2 = 0.277	Adjusted R2 = 0.466
Composite z-score
Continuous
NPS	−0.56 (−0.65 to −0.47), <.001	−0.31 (−0.40 to −0.23), <.001	−0.18 (−0.26 to −0.11), <.001
Model fit	Adjusted R2 = 0.025	Adjusted R2 = 0.284	Adjusted R2 = 0.438
Categorical
Low NPS	Ref	Ref	Ref
High NPS	−1.29 (−1.53 to −1.06), <.001	−0.70 (−0.90 to −0.49), <.001	−0.43 (−0.62 to −0.24), <.001
Model fit	Adjusted R2 = 0.003	Adjusted R2 = 0.276	Adjusted R2 = 0.436

Model 1: Non-adjusted model; Model 2 adjusted for age, gender, and race; Model 3 adjusted for age, gender, race, education, PIR, BMI, drinking, smoking, physical activity, and history of diseases (hypertension, heart disease, stroke, diabetes).

The symbol β denotes the regression coefficient, and R² denotes the coefficient of determination. The composite z-score was calculated by summing the z-scores ([test score − mean score]/SD) of the 3 individual tests.

AFT = Animal Fluency Test, BMI = body mass index, CERAD W-L = Consortium to Establish a Registry for Alzheimer’s Disease Word Learning subtest, CI = confidence interval, DSST = Digit Symbol Substitution Test, NPS = Naples prognostic score, PIR = poverty-income ratio, Ref = reference category, SD = standard deviation.

The multivariate adjusted odds ratios (ORs) and 95% CIs for the odds of cognitive impairment, analyzed in relation to continuous NPS values and stratified by NPS levels, as determined through a multiple logistic regression model, are shown in Table 4. When considered a continuous variable, the NPS demonstrated a significant association solely with the AFT score (OR = 1.16, 95% CI: 1.02–1.33, P = .021). Following adjustment for confounding covariates, the high-NPS group had higher odds of cognitive impairment than the low-NPS group according to both the AFT (OR = 1.41, 95% CI: 1.04–1.92, P = .027) and DSST (OR = 1.47, 95% CI: 1.00–2.14, P = .047). Nonetheless, no significant association was observed between the NPS and cognitive impairment defined by the CERAD W-L or the global cognitive z-score.

Table 4.

Logistic regression analyses for the association between NPS and cognitive impairment.

	Model 1	Model 2	Model 3
	OR (95% CI), P	OR (95% CI), P	OR (95% CI), P
CERAD W-L score
Continuous
NPS	1.25 (1.13–1.40), <.001	1.01 (0.90–1.13), .849	0.94 (0.83–1.07), .356
Categorical
Low NPS	Ref	Ref	Ref
High NPS	1.54 (1.18–2.02), .001	1.00 (0.75–1.32), .983	0.88 (0.65–1.19), .400
AFT score
Continuous
NPS	1.32 (1.18–1.48), <.001	1.22 (1.08–1.38), .001	1.16 (1.02–1.33), .021
Categorical
Low NPS	Ref	Ref	Ref
High NPS	1.94 (1.49–2.53), <.001	1.56 (1.17–2.08), .002	1.41 (1.04–1.92), .027
DSST score
Continuous
NPS	1.39 (1.24–1.57), <.001	1.26 (1.10–1.43), .001	1.16 (0.99–1.36), .069
Categorical
Low NPS	Ref	Ref	Ref
High NPS	2.18 (1.67–2.85), <.001	1.64 (1.20–2.25), .002	1.47 (1.00–2.14), .047
Composite z-score
Continuous
NPS	1.35 (1.20–1.51), <.001	1.14 (1.01–1.29), .040	1.04 (0.90–1.20), .616
Categorical
Low NPS	Ref	Ref	Ref
High NPS	2.01 (1.54–2.63), <.001	1.35 (1.01–1.81), .044	1.16 (0.83–1.62), .385

Model 1: Non-adjusted model; Model 2 adjusted for age, gender, and race; Model 3 adjusted for age, gender, race, education, PIR, BMI, drinking, smoking, physical activity, and history of diseases (hypertension, heart disease, stroke, diabetes).

The composite z-score was calculated by summing the z-scores ([test score − mean score]/SD) of the 3 individual tests.

AFT = Animal Fluency Test, BMI = body mass index, CERAD W-L = Consortium to Establish a Registry for Alzheimer’s Disease Word Learning subtest, CI = confidence interval, DSST = Digit Symbol Substitution Test, NPS = Naples prognostic score, OR = odds ratio, PIR = poverty-income ratio, Ref = reference category, SD = standard deviation.

3.4. Subgroup analysis

Subgroup analyses and interaction tests were performed to assess the consistency of the associations between a high NPS and cognitive impairment measured using the AFT and DSST after adjustment for all covariates (Fig. 2). For the AFT, stronger associations were observed among participants with a PIR < 1.3 (OR = 2.28, 95% CI: 1.31–3.96), males (OR = 1.75, 95% CI: 1.13–2.73), and those with higher education levels (OR = 1.66, 95% CI: 1.05–2.62). For the DSST, significant associations were observed among participants with a PIR < 1.3 (OR = 2.40, 95% CI: 1.25–4.59), those with higher education levels (OR = 2.39, 95% CI: 1.28–4.44), and participants aged 60 to 69 years (OR = 2.13, 95% CI: 1.06–4.26). Significant associations were also observed in some subgroups without diabetes or heart disease. However, no statistically significant interactions were identified for either outcome (all P for interaction > .05), indicating that the associations did not differ significantly across the examined subgroups.

Figure 2.

Subgroup analyses of the association between NPS and the Animal Fluency Test (AFT) scores (A), and subgroup analyses of the association between NPS and the Digit Symbol Substitution Test (DSST) scores (B). Models were adjusted for age, sex, race, education, PIR, BMI, drinking, smoking, physical activity, and history of diseases (hypertension, heart disease, stroke, diabetes). BMI = body mass index, CI = confidence interval, NPS = Naples prognostic score, PIR = poverty-income ratio.

4. Discussion

In this cross-sectional study, we used NHANES, a health survey database that is representative of the US population, and found a negative association between the NPS and cognitive function in older adults. Specifically, a higher NPS was linked to poorer performance on the AFT and DSST. Furthermore, even after controlling for potential confounding variables, the relationships remained statistically significant. Notably, the observed associations were particularly pronounced among certain high-risk groups, including individuals with a PIR < 1.3, those with higher education levels, and those without a history of diabetes mellitus or heart disease. These findings suggest that the NPS may provide supplementary information about cognitive performance in older adults, although its clinical utility requires further prospective evaluation. Our findings provide a basis for future prospective studies examining the mechanisms underlying these associations and evaluating the potential clinical utility of the NPS.

Notably, no significant association was observed between the NPS and CERAD W-L performance after full adjustment. Although the individual NPS components were significantly correlated with CERAD W-L in the unadjusted Spearman analyses, these component-level correlations differ from the adjusted association between the composite NPS and CERAD W-L. The CERAD W-L primarily assesses immediate and delayed verbal learning and memory,^[18] whereas the AFT and DSST capture verbal fluency, executive function, attention, and processing speed.^[19] This difference may indicate that the association between the NPS and cognitive function is domain-specific, with inflammatory and nutritional status being more closely related to executive and processing-speed functions than to verbal memory.^[23] In addition, CERAD W-L performance may be influenced by factors not fully captured in the present analysis, including underlying neurodegenerative pathology, educational background, and practice or ceiling effects.^[24–26] Therefore, the absence of a statistically significant association should not be interpreted as evidence that the NPS is unrelated to memory function. Longitudinal studies using more comprehensive memory assessments are needed to clarify this relationship.

Recent research suggests that the role of inflammation in the pathogenesis of neurodegenerative diseases, including AD, may surpass the traditionally emphasized theories centered on beta-amyloid and tau proteins. Neuroinflammation has been identified as an early contributor to AD development.^[27] Furthermore, a significant correlation exists between markers of systemic inflammation and the severity of AD.^[28] A meta-analysis demonstrated that an increased NLR is associated with greater cognitive decline in individuals.^[10] A separate study indicated that the NLR was markedly elevated in patients over 60 years of age with cognitive impairment, which was correlated with an increased risk of developing dementia over a 5.9-year follow-up period.^[29] Research by Li et al identified LMR values as indicators of immunological alterations in the brains of dementia patients, with lower LMR values being linked to cognitive decline.^[30] In the component-level Spearman analyses, a higher NLR and a lower LMR were associated with lower CERAD W-L and composite z-scores, whereas their correlations with AFT and DSST were not statistically significant. These findings suggest that the relationships of NLR and LMR with cognitive performance may differ across cognitive domains.

Nutritional status serves as a significant indicator of brain health and cognitive function. A growing body of evidence has indicated a correlation between ALB levels and cognitive impairment.^[31,32] In addition, cholesterol is considered a vital factor in the development and progression of AD. Huan et al reported consistent positive associations between total serum cholesterol and cognitive performance.^[33] In our component-level analyses, higher ALB and TC levels were positively correlated with CERAD W-L, AFT, DSST, and composite z-scores. These findings indicate that lower ALB and TC levels were associated with poorer cognitive performance across the cognitive measures examined.

The NPS represents an innovative, multidimensional assessment system that evaluates inflammatory and nutritional status, incorporating the NLR, LMR, ALB level, and TC. Recent investigations have demonstrated that the NPS exhibits enhanced prognostic predictive capacity in various populations. For example, the NPS is significantly associated with adverse outcomes in patients with metabolic syndrome, cardiovascular disease, nonalcoholic fatty liver disease, chronic kidney disease, and respiratory diseases, thereby demonstrating its potential as a prognostic indicator.^{[13–15,34–38]} A recent study also reported that an elevated NPS was independently associated with cognitive impairment in patients undergoing hemodialysis.^[16] This finding suggests that the NPS may provide supplementary information for identifying patients with cognitive impairment. In the present study, our results were also in alignment with previous studies; both linear and logistic regression analyses indicated that a higher NPS was associated with poorer cognitive performance. Moreover, when cognitive function was analyzed as a binary outcome (impaired vs nonimpaired) via logistic regression, a higher NPS was significantly associated with cognitive impairment according to the AFT and DSST. These findings indicate cross-sectional differences in cognitive performance and do not establish within-person cognitive change over time.

Interestingly, when we analyzed the data using linear regression models, we found a significant negative correlation between the NPS and global cognitive z-scores (β = −0.18, 95% CI: −0.26 to −0.11, P < .001), indicating that participants with higher NPS values had lower overall cognitive function. This demonstrates a broader association between the NPS and cognitive performance. However, when using the global cognitive z-score as the outcome in logistic regression Model 3, the association with NPS was not statistically significant. This may be partly explained by the relationships between cognitive function and other covariates. Li et al revealed that BMI and weight-adjusted-waist index were associated with cognitive function decline in older participants,^[39] while Wang revealed that smoking was another factor associated with cognitive decline.^[40] Moreover, in patients with hypertension, probiotic intake was linked to a higher global cognition z-score.^[41] Collectively, this evidence suggests that adjustment for covariates may attenuate associations with cognitive outcomes and that further investigation is warranted. The difference in findings between the linear regression model and the binary logistic regression model could be explained by the differing natures of the 2 types of analyses. Linear regression retains the full range of continuous cognitive scores and may therefore detect differences that are obscured when cognitive performance is dichotomized for logistic regression. As a result, the lack of a significant association in the binary regression model does not necessarily refute the association between the NPS and global cognitive function observed in the linear regression model.

However, the mechanisms underlying the association between the NPS and cognitive performance should be further explored. The NPS was calculated based on the NLR, LMR, ALB level, and TC, and these laboratory examinations have been widely researched in previous studies. He et al found that a higher baseline NLR predicted worsening Letter–Number Sequencing scores and was negatively correlated with Montreal Cognitive Assessment scores.^[42] Xiao et al reported that a higher NLR was associated with more rapid cognitive decline in patients with Parkinson disease in the UK Biobank dataset.^[43] LMR has also been investigated as a predictor of inflammation-related conditions. Hu et al reported that a low preoperative LMR may help identify patients at high risk of postoperative cognitive dysfunction.^[44] However, Mehta et al found that NLR, but not LMR, was associated with greater longitudinal cognitive decline,^[11] suggesting that the associations of NLR and LMR with cognitive function warrant further investigation. The association between TC and cognitive function has been studied for decades. van den Kommer et al reported that lower TC may be considered as a frailty marker of lower cognitive function in older adults.^[45] A more recent study indicated the association between incremental TC and a slower annual decline in Mini-Mental State Examination scores.^[46] ALB is a marker of general health and may also be relevant to cognitive function. Si et al reported higher glycated ALB levels in patients with type 2 diabetes mellitus and cognitive impairment. Zahr and Pfefferbaum showed that participants with higher ALB may be associated with white matter hyperintensities, partially explaining the anatomical basis of cognitive decline.^[47,48] Taken together, previous evidence and the present component-level analyses suggest that the individual NPS components may capture different aspects of inflammatory and nutritional status associated with cognitive performance. The NPS combines these components into a single composite measure.

Subgroup analyses identified significant associations between a higher NPS and cognitive impairment according to the AFT and DSST in several subgroups. However, none of the interaction tests were statistically significant. Although the point estimates varied across subgroups, the present findings do not establish differences in the association according to physiological or socioeconomic characteristics. Lower PIR may be associated with differences in access to nutritional care and healthcare, as well as with inflammatory, nutritional, and cognitive measures. On the other hand, higher education may be associated with greater cognitive reserve and differences in measured cognitive performance.

The clinical implications of this study are significant. The findings highlight the importance of assessing both nutritional and inflammatory status in older adults as part of a comprehensive approach to managing cognitive health. Given that the NPS is a simple, composite score based on routine clinical markers (ALB, cholesterol, NLR), it may provide supplementary information when determining whether further cognitive assessment is warranted. However, the present cross-sectional findings do not establish that interventions targeting the components of the NPS would improve cognitive outcomes.

Although our research offers important insights into the association between the NPS and cognitive function, it has several limitations. First, the cross-sectional nature of the analysis limits the ability to draw causal conclusions. We observed an association between a higher NPS and poorer cognitive function; however, large-scale prospective studies are needed to determine the temporal relationship between the NPS and changes in cognitive function. Second, we relied on a limited set of cognitive tests (AFT, DSST, CERAD), and future studies should incorporate a more comprehensive battery of neuropsychological assessments to assess cognitive function more robustly. Third, the study did not include additional biomarkers of inflammation and nutrition, such as C-reactive protein or the ALB–globulin ratio, which could offer a more detailed understanding of the underlying mechanisms. Finally, future prospective studies may use mediation analysis to better understand the relationships among these factors and cognitive function.

5. Conclusion

A higher NPS was associated with poorer performance in verbal fluency, processing speed, and global cognitive function among older adults. Because the NPS is calculated from routinely available inflammatory and nutritional indicators, it may provide clinicians with supplementary information when identifying older adults who may benefit from further cognitive assessment. However, the NPS should not be used as a stand-alone screening or diagnostic tool based on the present cross-sectional findings. Future prospective studies should determine whether the NPS predicts longitudinal cognitive change, validate clinically useful cutoff values, and assess whether its addition to existing cognitive risk models improves risk discrimination and clinical decision-making.

Acknowledgments

The dataset was obtained from the NHANES website, which is supported by the CDC and is accessible to the public for free.

Author contributions

Conceptualization: Ke Wei.

Data curation: Ying Xiong, Jin Fang.

Formal analysis: Ke Wei.

Investigation: Ying Xiong, Ke Wei, Jin Fang.

Resources: Ying Xiong, Jin Fang.

Software: Ke Wei.

Supervision: Ying Xiong.

Visualization: Jin Fang.

Writing – original draft: Ying Xiong, Ke Wei, Jin Fang.

Writing – review & editing: Ying Xiong, Ke Wei, Jin Fang.