Hypoalbuminemia and Osteoporosis: Reappraisal of a Controversy

Abstract

Context:

Human studies have reported conflicting results on the association of hypoalbuminemia with osteoporosis.

Objective:

The aim of the study is to test the independent association between hypoalbuminemia and osteoporosis.

Design:

This is a cross-sectional observation.

Setting and Participants:

Patients are the outpatient consecutive individuals with available clinical, laboratory, and densitometry data from 2001 to 2013 in our tertiary care academic medical center.

Exposure:

Exposure is hypoalbuminemia defined as serum albumin less than 3.5 g/dL.

Main Outcome Measure:

Osteoporosis is defined as bone mineral density of 2.5 SD or less below the mean peak bone mass of young, healthy adults.

Results:

Overall, 21 121 patients were included. Mean of age was 61 years (SD 14). There were 4244 males (20.1%) and 1614 patients of African-American ethnicity (7.6%). There was a graded decrease in rate of osteoporosis from 28.0% (n = 33) at albumin of 3 g/dL or less to 9.3% (n = 1548) at albumin greater than 4 g/dL (P < .001) at the femoral neck and from 20.3% (n = 24) to 6.1% (n = 1011) at the total hip (P < .001). In a fully adjusted model, the odds of osteoporosis at albumin of 3 g/dL or less was 3.31-fold (95% confidence interval [CI] 2.08–5.28, P < .001) at the femoral neck, 2.98-fold (95% CI 1.76–5.01, P < .001) at the total hip, and 2.18-fold (95% CI 1.43–3.31, P < .001) at the lumbar spine as compared with albumin greater than 4 mg/dL. A similar independent association was identified with a longer-observed duration of hypoalbuminemia.

Conclusion:

In a large population, we report an independent association of osteoporosis with lower levels of serum albumin and a longer-observed duration of hypoalbuminemia.

Osteoporosis (OP) is a public health issue with a prevalence up to 38% in women in industrialized countries, affecting up to 49 million people (1). It is associated with hip and vertebral fractures (2, 3) and is linked to higher mortality (4–7). Hypoalbuminemia is commonly observed in hospitalized patients with reported incidence as high as 60% of inpatients (8–10). In addition to the well-recognized physiological roles of albumin in the maintenance of oncotic pressure and compound transportation, its alteration is also linked to a flurry of systemic disorders such as cancer, liver disease, nephrotic syndrome, chronic inflammation, and malnutrition (11, 12). Despite vast literature in the context of systemic disorders, the association of hypoalbuminemia with bone mineral density (BMD) and OP has been a matter of controversy. Although few observations have suggested presence of such a link (13–15), some others have reported the lack of an independent association (16, 17). Such discrepancy in the literature is a reflection of differences in study population, a limitation in the diversity of selected participants, variation in design, limitations in sample size and power of study, suboptimal statistical analysis, and a limited range in which serum albumin has been studied. We hypothesize that hypoalbuminemia is independently associated with OP at the lumbar spine and hip sites. We sought to examine the following: 1) whether serum albumin levels associate with OP, 2) whether duration of hypoalbuminemia correlates with OP, and 3) whether changes of BMD over time track serum albumin.

Materials and Methods

This study is a cross-sectional observation of all included patients and a longitudinal analysis of the subgroup of patients with multiple densitometries. We obtained the institutional review board approval (HUM00075043). Because the study is an observation of existing data sets, a need for informed consent was waived by the institutional review board. Description of patient population and the method of data collection have been published elsewhere (18). In brief, the population is defined as individuals with bone densitometry as outpatients at the University of Michigan Health System from August 2001 to October 2013. Inclusion criteria were availability of pertinent key variables including demographics, laboratory values, use of medications, and history of comorbidities. Exclusion criteria were age younger than 18 years, and chronic kidney disease (CKD) stages 4 and 5 (estimated glomerular filtration rate [GFR] < 30 mL/min). The rationale for excluding patients with advanced CKD was to enrich the patient population having low BMD with OP as opposed to renal osteodystrophy. All the consecutive patients who met the inclusion and exclusion criteria were included in the analyses.

Data on BMD were obtained from the server of the densitometry unit at the University of Michigan Health System. Demographic variables were age, sex, race, weight, and height at the time of densitometry. Laboratory variables including plasma glucose, serum albumin, sodium, calcium, and creatinine, use of medications known to impact BMD (Supplemental Table 1), and history of comorbidities (Supplemental Table 2) known to impact BMD using International Classification of Diseases, ninth revision, codes were retrieved from electronic medical records. The uses of medications, history of comorbidities, and laboratory values for the past 2 years prior to the date of the last densitometry were retrieved. For the subset of patients with multiple densitometries, the last examination was used for the cross-sectional analysis due to the higher likelihood of having OP contributing to higher study power. The subset of patients with more than one densitometry underwent further longitudinal analysis using the values from their first and their last densitometries for the downstream analyses. Time-averaged serum albumin for each patient was calculated from the values of the measured serum albumin within the past 2 years prior to the date of the last densitometry. Hypoalbuminemia was defined as serum albumin less than 3.5 g/dL (<35 g/L). Observed duration of hypoalbuminemia was defined as the sum of the intervals between two successive hypoalbuminemic values. We defined OP according to the definition of World Health Organization as a BMD of 2.5 SD or less below the mean peak bone mass of young, healthy adults (19). A CKD-Epidemiology Collaboration formula was used to calculate GFR (20). General Electric Lunar iDXA densitometers were used for densitometry studies. The intraobserver coefficient of variation at lumbar and hip anatomical areas was 2.39% and 1.65%, respectively (18).

Statistical analysis

We used mean and SD for description of normally distributed variables and frequency and percentage for description of categorical variables. χ² was used to test the association of categorical variables in different groups. ANOVA with Bonferroni post hoc correction for multiple comparisons was applied to compare mean of continuous variables between independent groups. Logistic regression analysis with forward stepwise iterative inclusion of predictors of OP was used to test the independent association of OP with categories of serum albumin or with observed duration of hypoalbuminemia. For that purpose, three different series of models were tested. Model 1 was case-mixed unadjusted model. Model 2 was adjusted with age, sex, race, and categories of body mass index (BMI). Model 3 was adjusted with components of model 2 plus smoking, other laboratory values (plasma glucose, serum sodium, calcium, and estimated GFR), medications (Supplemental Table 1), and comorbidities (Supplemental Table 2). Analysis of covariance was used to compare the changes of BMD over time between categories of serum albumin. The covariates that the analysis of covariance model was adjusted for were the variables that could significantly and independently explain the differences in interval change of BMD over time and included age, BMI, interval between the two densitometries, history of solid organ transplant, bariatric surgery, and use of bisphosphonates, loop diuretics, and aromatase inhibitors.

The study has greater than 93% power to detect the smallest observed odds ratio (OR) at albumin of 3 g/dL or less (≤30 g/L) (see Figure 2A), greater than 98% at an albumin level greater than 3 to less than 3.5 g/dL(>30 to < 35 g/L), and greater than 98% at an albumin level of 3.5–4 g/dL (35 to 40 g/L) using a two-sided test at an alpha level of .01 and an albumin greater than 4 g/dL (>40 g/L) as the reference category. Statistical Package for the Social Sciences (SPSS) version 22 was used for the analyses.

Figure 2.

A, Comparison of the OR of osteoporosis by categories of time-averaged serum albumin shows a significantly higher odds of osteoporosis at all subgroups compared with albumin greater than 4 g/dL (>40 g/L, reference), at all different anatomical sites, in all three models (P < .001). B, Model 1: unadjusted; model 2: model 1 plus age, sex, race, and BMI; model 3: model 2 plus smoking, medications, comorbidities, and laboratory values including serum sodium, calcium, plasma glucose, and GFR. Odds ratio ± 95% CI is presented. To convert values of serum albumin to grams per liter, multiply by 10.

Results

Baseline

We included 21 121 patients in this study. The mean age was 61 years (SD 14) with a range from 20 to 103 years. There were 4244 males (20.1%) and 1614 patients of African-American ethnicity (7.6%). Overall 131 817 measures of serum albumin from the selected patients within a 2-year time frame before the date of the last densitometry contributed to the available data on serum albumin. The median number of measured serum albumin per patient was 2 with an interquartile range (IQR) of 1–5 and a range of 1–234. The interval between the date of the last measured albumin and the date of the densitometry had a median of 71 days and an IQR of 21–224 days. The mean of time-averaged serum albumin was 4.24 ± 0.35 g/dL (42.4 ± 3.5 g/L, Figure 1A). The distribution of baseline characteristics by categories of time-averaged serum albumin is shown in Table 1. There was a significant increase in proportion of females by categories of serum albumin. Whereas there was a statistically significant decreasing trend in the rate of use of loop diuretics with increased serum albumin, the other medications showed a significant increase in the trend (P < .001). Similarly, there was a significant decrease in some of the comorbidities with increasing serum albumin (P < .001).

Figure 1.

A, Distribution of time-averaged serum albumin in the study population. B, Comparison of the crude rate of osteoporosis by categories of time-averaged serum albumin shows a significantly higher rate of osteoporosis at all subgroups compared with albumin greater than 4 g/dL (>40 g/L) and at all different anatomical sites (P < .001). Bars are a percentage plus SE. To convert values of serum albumin to grams per liter, multiply by 10.

Table 1.

Comparison of Distribution of Baseline Characteristics of the Study Patients by Categories of Time-Averaged Serum Albumin^a

	≤3 g/dL	>3 to 3.4 g/dL	3.5–4 g/dL	>4 g/dL
n	118	597	3798	16 608
Age, y	57 ± 18	62 ± 17	63 ± 16	61 ± 14
Female gender, %b	63.6	72.7	75.2	81.4
BMI, kg/m2	25.8 ± 5.5	26.8 ± 6.5	27.9 ± 7.0	27.5 ± 6.2
Smoking, %b	19.5	26.1	30.8	36.6
Ethnicity
Caucasian, %	82.2	82.9	82.7	84.3
Asian, %	2.5	2.5	3.9	6.1
African American, %	13.6	12.6	10.3	6.8
Others, %	1.7	2.0	3.1	2.8
Medications
Bisphosphonates, %b	4.2	8.2	9.8	12.5
Steroids, %c	23.7	27.5	30.3	27.6
Thiazides, %b	5.1	9.5	12.6	16.3
Loop diuretics, %b	22.9	12.6	9.4	4.9
Antiepilepticsc	8.5	10.6	11.6	12.9
Antidepressantsb	15.3	17.6	21.4	26.8
Opiatesc	21.2	27.5	27.0	28.5
Heparinsb	7.6	7.7	5.3	2.4
Medical history
Bariatric surgery, %b	6.8	11.2	9.5	6.6
Malabsorption, %b	5.1	2.8	1.8	1.1
Solid organ transplant, %b	9.3	10.6	8.9	4.1
Liver cirrhosis, %b	43.2	26.0	18.9	15.7
Heart failure, %b	28.0	26.8	20.4	10.5
Plasma glucose, mg/dL	106 ± 44	112 ± 47	107 ± 41	103 ± 31
Calcium, mg/dL	8.7 ± 0.7	9.0 ± 0.6	9.2 ± 0.5	9.6 ± 0.5
Sodium, meq/L	139 ± 3	140 ± 3	140 ± 3	140 ± 2
GFR, mL/min	84 ± 29	78 ± 26	77 ± 23	80 ± 20
Lumbar L1-L4 BMD, g/cm2	1.10 ± 0.20	1.12 ± 0.22	1.15 ± 0.21	1.14 ± 0.19
Femoral neck BMD, g/cm2	0.83 ± 0.17	0.83 ± 0.17	0.87 ± 0.16	0.88 ± 0.15
Total hip BMD, g/cm2	0.87 ± 0.18	0.87 ± 0.17	0.91 ± 0.17	0.93 ± 0.15

To convert values for glucose to millimoles per liter, multiply by 0.05 551. To convert values of calcium to millimoles per liter, multiply by 0.25. To convert vales of sodium to millimoles per liter, multiply by 1.

^aValues are mean ± SD or count and percentage.

^bP < .001.

^cP < .05.

Rate of osteoporosis by categories of serum albumin and anatomical sites

Figure 1B illustrates that there was a significant graded decline in the crude rate of OP at the lumbar spine from 31.4% (n = 37) at an albumin of 3 g/dL or less (≤30 g/L) to 18.4% (n = 3057) at an albumin greater than 4 g/dL (>40 g/L, P < .001). Similarly, there was a statistically significant decline in the crude rate of OP at the femoral neck from 28.0% (n = 33) at an albumin of 3 g/dL or less (≤30 g/L) to 9.3% (n = 1548) at an albumin greater than 4 g/dL (>4 g/L, P < .001) and a significant decline at the total hip area from 20.3% (n = 24) to 6.1% (n = 1011) at the corresponding categories of albumin (P < .001).

Risk of osteoporosis by categories of time-averaged serum albumin

Figure 2A shows that there was a graded increase in odds of OP by decrease in level of serum albumin as compared with an albumin greater than 4 g/dL (>40 g/L, reference level) using an unadjusted model (model 1), showing a significantly higher OR at all categories of serum albumin compared with the reference level at all anatomical sites (P < .001). The significance of the OR remained unchanged after adjusting for age, sex, race, and BMI (P < .001, model 2) or with additional adjusting by medications, comorbidities, and laboratory values in addition to components of model 2 (P < .001, model 3) at all subgroups of serum albumin and at all anatomical sites.

Risk of osteoporosis by observed duration of hypoalbuminemia

In addition to testing the association of OP with different levels of time-averaged serum albumin, we tested the association of OP with the quartiles of observed duration of hypoalbuminemia. Patients with normal-appearing, time-averaged serum albumin might have fewer measures of serum albumin at the hypoalbuminemic range, but due to having more measures at normal range, their corresponding mean value could fall in the normal range. As a result the number of patients with at least one hypoalbuminemic interval exceeds the number of patients whose time-averaged serum albumin was less than 3.5 g/dL (<35 g/L). We identified 1743 patients with at least one hypoalbuminemic interval, including 405 patients at the highest quartile of the observed duration of hypoalbuminemia, 425 patients in the third, and 449 patients in the second, and the rest of the other patients were in the first quartile. We aggregated the patients with no hypoalbuminemia with the first quartile. The median interval between the first measure of serum albumin and the last densitometry was 19.9 months in the first quartile, 18.9 in the second, 21.0 in the third, and 21.9 months in the fourth quartile, showing equally distributed intervals independent of the observed duration of hypoalbuminemia, suggesting that the clinical decision to measure serum albumin and to perform densitometry were independent of each other, making detection bias unlikely. The 25th, 50th, and 75th percentiles of the hypoalbuminemic interval were 5, 28, and 122 days, respectively. When compared with the first quartile, the greatest odds of association with OP was noted at the longest observed duration of hypoalbuminemia at all anatomical sites (P < .001, Figure 2B). Adjusting by demographics and BMI (model 2) or full adjustment (model 3) did not change the association (Figure 2B).

Change of BMD by serum albumin in longitudinal cohort

A subset of the cohort included 7896 patients who had more than one densitometry with available serum albumin within the past 2 years prior to the last densitometry. The median number of densitometries was 2 with an IQR of 2–3. Compared with patients with a single densitometry, those with multiple densitometries on average were older, had a higher BMI, and were more likely to have other comorbidities and had higher rate of medication use (Supplemental Table 3). The median interval between the two densitometries was 55 months with an IQR of 34–82 months. There were 125 patients with time-averaged serum albumin less than 3.5 g/dL (35 g/L), 1062 with albumin between 3.5 and 4 g/dL (35–40 g/L), and 6709 with serum albumin of greater than 4 g/dL (40 g/L). Figure 3 shows that at the femoral neck and total hip areas, the BMD of the patients at the last densitometry had significantly decreased as compared with the first examination at all categories of time-averaged serum albumin (P < .001). The difference between the first and the last BMD did not reach to statistical significance at lumbar spine. Similarly, the increasing linear trend of BMD between the categories of serum albumin at the first as well as the last examination was highly significant at both the femoral neck and total hip areas (P < .001) but not at the lumbar spine. Because the characteristics of the patients with single and multiple densitometries were different from the list of background variables (including demographics, anthropometrics, laboratory values, medications, and comorbidities), we identified the variables that could partially and independently explain the change of BMD over time between different categories of albumin, which included age, BMI, interval between the two densitometries, history of solid organ transplant, bariatric surgery and use of bisphosphonates, loop diuretics, and aromatase inhibitors. After adjusting for these covariates, we show that the decrease of BMD at the femoral neck at albumin less than 3.5 g/dL (35 g/L) was significantly more than what was observed at an albumin level of 3.5–4 g/dL (35–40 g/L, P = .005) and greater than 4 g/dL (> 40 g/L, P < .001, Figure 4) independent of other variables. Similarly, the decrease of BMD at the total hip at albumin less than 3.5 g/dL (35 g/L) was significantly more than what was observed at an albumin level of 3.5–4 g/dL (35–40 g/L, P = .018), or greater than 4 g/dL (P < .001), after similar adjustments. Such a decline did not reach to statistical significance at lumbar spine.

Figure 3.

Comparing the first (gray color) and the last (black color) mean BMD of the patients with multiple densitometries by categories of time-averaged serum albumin at different anatomical sites. The difference of the BMD between the first and the last densitometry within each category of serum albumin in the femoral neck and total hip reached statistical significance (***, P < .001). The difference is not significant at subgroups of the lumbar spine. Similarly, the increasing trend of BMD between albumin categories at the first and last densitometries is significant at the femoral neck and total hip (P < .001) but not at the lumbar area. Bars are percentage plus SE. NS, not significant. To convert values of serum albumin to grams per liter, multiply by 10.

Figure 4.

Comparing change of BMD from the first to the last densitometry after adjusting for the age, BMI, interval between the two densitometries, history of bariatric surgery, solid organ transplantation, and use of bisphosphonates, loop diuretics, and aromatase inhibitors by categories of time-averaged serum albumin at different anatomical sites. *, P < .05, **, P < .01, ***, P < .001 compared with albumin of 3.5 g/dL or less (≤35 g/L). Values are mean of BMD change ± SE. To convert values of serum albumin to grams per liter, multiply by 10.

Subgroup analysis

Figure 5 shows that the OR (95% confidence interval [CI]) of the association of hypoalbuminemia with OR has remained highly significant at all subgroups of age, gender, race, and BMI at the femoral neck and total hip and most subgroups in the lumbar spine, except in overweight subgroups at the lumbar spine (OR 1.228, 95% CI 0.857–1.760, P = .263).

Figure 5.

Comparing the OR (95% CI) of the association between hypoalbuminemia (serum albumin < 3.5 g/dL) with osteoporosis by subgroups of age, gender, race, and BMI. AA, African-American.

Discussion

In this study we observe an inverse relationship of serum albumin level with prevalent OP. The association of lower levels of serum albumin with OP was highly dose responsive, highly significant, and independent of any other variables at all anatomical sites. Aligned with this observation, we also noticed an independent, highly significant association between OP and observed duration of hypoalbuminemia, with a trend toward higher risk of OP at the longest observed duration of hypoalbuminemia. In a subset of patients with multiple densitometries, a decrease of BMD over time was greater in patients with time-averaged serum albumin less than 3.5 g/dL (35 g/L) as compared with others at the femoral neck and total hip independent of other covariates associated with a change of BMD between the two densitometries.

The importance of albumin in pathogenesis of OP in human was first suggested by Albright and Reifenstein in 1947 (13). Later on, an animal model experiment showed that the rats that were fed a low albumin diet for 3 months had a significantly lower BMD as compared with those on normal albumin diet (21). In the same study, the investigators used the Nagase analbuminemic rat (a mutant strain with serum albumin deficiency established by Nagase et al [22]) and showed that as compared with other groups, the ovariectomized Nagase rats had a significantly lower BMD at 14 and 16 weeks in their proximal tibia but not in their total BMD (21). The results from human studies are more conflicting. In a case-control study Rico et al (14) compared the characteristics of 25 osteoporotic elderly women with 20 healthy women and reported a significantly lower serum albumin level in the former group. Similarly, in an analysis of 217 healthy individuals, Saito et al (15) showed an association between a decreased peak BMD and low albumin. More recently the impact of liver failure as an etiological factor on mineral metabolism was called to attention in two studies (23, 24). In one study Tsuneoka et al (23) showed a decreased lumbar spinal BMD in 60 patients with liver cirrhosis and chronic viral hepatitis as compared with 40 healthy sex- and age-matched individuals. Similarly, Diamond et al (24) reported decreased mean trabecular bone volume and thickness in various forms of liver diseases along with a reduction of osteoblastic surface.

Contrary to these findings, two studies have questioned the independent association of serum albumin with BMD (16, 17). The Rancho Bernardo Study reported a weak positive correlation between serum albumin and BMD, which disappeared after adjusting for age (17). Similarly, in a group of healthy postmenopausal women, no correlation was found between serum albumin and BMD (16). All the human observational studies have suffered from methodological limitations including relatively low sample size (14, 15, 23, 24), lack of multivariable adjustment (14–16, 23), lack of age, gender or race diversity (14–17), and narrow range of studied serum albumin (17). More specifically in the Rancho Bernardo study, only 4 of 1593 values of serum albumin were outside the normal range of 3.5–5.0 g/dL (35–50 g/L) (17). As a result, the variability of BMD within the normal range of serum albumin was best described by age; hence, serum albumin did not have sufficient variability to include adequate number at hypoalbuminemic range necessary to capture its independent association with BMD. Similarly, the narrowly defined selection criteria to include only healthy postmenopausal elderly women have likely eliminated the variability necessary to capture the aforementioned associations in the other negative study (16). Our study is in agreement with the previous literature, which has suggested an association between hypoalbuminemia and OP (13–15).

Our study has several strengths that make it the best available evidence of the association to date. To our knowledge this is the largest study in the field exploring the relationship. It included a wide age range (20 to > 100 y), is sufficiently gender and race inclusive, and is adequately powered to test the proposed hypotheses. The large sample size allowing adjustment for a large number of potential confounders and granular available clinical and laboratory data have allowed efficient modeling aimed at assessing the independence of the associations. Because the results from unadjusted to fully adjusted models were similar, it was unlikely that the high number of covariates in the fully adjusted model influenced the associations. Furthermore, it signifies that the association of hypoalbuminemia with low osteoporosis was independent of any potential confounders. Novel aspects of this observation are the availability of multiple measures of serum albumin, which has allowed testing not only the dose-response relationship with different categories of serum albumin but also with its duration. The fact that the largest decline of BMD was observed at the lowest category of serum albumin independent of the predictors of BMD change is suggestive of an etiological rule for hypoalbuminemia in the pathogenesis of OP.

The mechanism(s) underlying the association between hypoalbuminemia with OP is unclear. Hypoalbuminemia may directly be linked with nuclear factor-κB (NF-κB), known to activate osteoclasts and suppress osteogenesis, or as an acute phase reactant to be indirectly linked with NF-κB via its association with inflammatory cytokines such as receptor activator of NF-κB ligand, TNFα, lymphotoxin, bacterial endotoxins, Toll-like receptor ligands, CD40L, IL-1, or oxygen radicals (25–27). Prolonged immobility is also known to cause diminished BMD (28). However, immobility is unlikely to be the only explanation of the observed association because the association is observed in a wide range of study populations. Albumin is present throughout osteoid and mineralized bone matrix, evidenced by its diffuse labeling in immunohistochemistry studies (29). Therefore, other mechanisms may include a decreased deposition of albumin in and increased efflux of albumin from spongiosal components of bone, decreased affinity for formation of calcium phosphate apatite crystals, alteration in homeostasis of parathyroid and vitamin D binding protein, increased sensitivity of bone to estrogen deficiency in women, preferential commitment of mesenchymal stromal cells toward the formation of adipogenic phenotype at the expense of osteogenesis, and decreased Gla protein with a net effect of decreased osteoblastic activity and increased osteoclast activity (21, 24, 30–32).

This study also has limitations. The greatest limitation is the retrospective nature of the study. It does not provide any mechanistic explanation for the observed associations. By design, this is an observational study, which does not allow inferring causal relationship. However, the significant strength of associations independent of any other risk factor, the dose-response relationship, and the stronger association not only with intensity of hypoalbuminemia but also with its longer duration necessitate further studies to explore possibility of potential causal mechanistic pathways. Although this is an observation of a large patient population, the results may not be extrapolated to other patient populations with different characteristics. Only a subset of our patients had more than one densitometry; however, the subset is sufficiently large and is adequately adjusted for relevant covariates to allow deriving meaningful conclusions. There may be additional unmeasured residual confounders, but because the results from the unadjusted and fully adjusted models were similar, the potential for presence of further residual confounders capable of drastically changing the results is low. Menopausal status was not recorded consistently in our database; however, because there was no inconsistency in the results by subgroup of gender and age, it is unlikely that menopausal status had impacted the observed associations. Sick patients were more likely to have more frequent measure of serum albumin, which could be a source for ascertainment bias. We believe that because the median interval between the first albumin measurement and the last densitometry was not clinically different by quartiles of hypoalbuminemic interval, the surrogate marker of observed duration of hypoalbuminemia is likely a valid estimate of the true duration of hypoalbuminemia, and the net effect of more frequent measures of serum albumin rather translates to increased precision of observed duration of hypoalbuminemia in sicker patients as opposed to change of the relationship between hypoalbuminemia and OP, and therefore, potential for ascertainment bias is unlikely to be the dominant explanation of the observed associations. This observation discloses a novel association with an important clinical implication. Although we did not show any mechanistic link between hypoalbuminemia and OP, the possible presence of such mechanistic pathways may expand our understanding of the mineral metabolism above and beyond known traditional markers and provide the opportunities for novel therapies and therapeutic targets. Further research is required to help better understand the pathophysiological link between albumin metabolism and OP.

Conclusion

We conclude that hypoalbuminemia was independently associated with a risk of OP at the femoral neck, total hip, and lumbar spine in a large patient population. Similarly, longer duration of hypoalbuminemia was independently associated with a higher risk of OP at the same anatomical sites. These findings provide perspectives regarding less clear novel aspects of bone metabolism.