One in 4 adult users of health services and the majority of older adults have multiple chronic conditions (MCC).1,2 Because of the paucity of information and the significant disease burdens from MCC, the 2010 US Department of Health and Human Services strategic framework called for an expansion of research on the disease trajectories over time for patients with comorbidities.3 Few studies describe the natural progression and clustering of MCC in all age groups.4 Among the leading causes of disabilities worldwide, 3 chronic conditions stand out due to their prevalence and high disease burden to individuals and society: obesity, hypertension, and depression. Obesity accounts for 5%–15% of deaths each year in the United States.5,6 Alone, it contributes to untimely mortality and decreased quality of life,7 both of which are further influenced by the frequent co-occurrence of debilitating comorbidities including depression, diabetes, and cardiovascular diseases.7–12
We chose hypertension and depression as comorbidities of obesity for 3 reasons. First, depression and hypertension are among the 14 priority conditions recognized by the Medicare Modernization Act Section 1013. Second, individually, each is among the leading causes of morbidity, mortality, and disability. Third, these 3 conditions often co-occur and make weight management more complicated, due to shared pathophysiological pathways13,14 and their impact on each other.13,15,16 Various combinations of these conditions frequently co-occur, and one can have synergistic effects on the development or outcomes of the others. For example, patients with depression gain weight17 faster compared with those without depression. Not only could depression cause obesity,13 treatment for depression can also worsen obesity.18 For instance, some antidepressant medication can make patients feel fatigued. Increased feelings of fatigue could reduce patient’s motivation to engage in behavioral modifications for weight and glucose management. In addition, many psychotropic medications could stimulate appetite, making weight management more challenging; a common side effect of some psychotropic medications is weight gain.18 In addition, it has been well documented that obesity is a risk factor for hypertension10 and depression.14
This study also demonstrates a meaningful use of electronic health records (EHRs) in a large community-based multispecialty practice. Investigators with access to just administrative data are constrained in their ability to examine temporal changes in diseases due to the lack of comprehensive data needed for such a study. Administrative data rely on coding to justify billing—if a visit occurs for an acute problem, a physician may note the patient’s blood pressure, but not enter a code for hypertension. Likewise, for many patients who are overweight, that was unlikely to be their reason for the visit. The advent of EHRs provides an opportunity to explore longitudinal changes in MCC in ways not previously possible.
Four research questions were addressed. First, for patients with high body mass index (BMI), how did co-occurring depression and/or hypertension affect the probability of having a diagnosis of overweight/obesity in the EHR? Second, for patients with high blood pressure (BP), how did having depression and/or high BMI affect the probability of having a diagnosis of hypertension recorded in the EHR? Third, how did having depression and/or hypertension influence BMI trajectories over an 8-year period? And lastly, how did having high BMI and/or depression affect BP trajectories, accounting for the propensity of being treated for hypertension? The Palo Alto Medical Foundation Institutional Review Board approved the study.
Study Population and Setting
This observational study focused on patients in one of the 3 geographic divisions of a large multispecialty group practice in California (the Group) between 2003 and 2010. Patients met all of the following inclusion criteria: (1) ≥2 in-person visits in primary care, (2) age 18 years or older in the first year of their observation, (3) ≥2 BMI measurements (for the analysis of patients with high BMI), (4) ≥2 recorded BP measurements (for the analysis of patients with high BP), and (5) met diagnostic or treatment criteria (described below) for either hypertension, overweight/obesity, or depression. Diagnostic criteria were derived from a problem list or encounter diagnoses. More than 90% of the diagnoses were based on the problem list only or on both the problem list and encounter diagnoses in the EHR. The remaining 10% were based on encounter diagnoses only. A total of 161,196 patients met these criteria. All the data on these 3 conditions were obtained from the EpicCare EHR system used by the Group.
Overweight or obese patients were identified by: (1) ICD-9 diagnosis codes for overweight/obesity (278.0, 278.00, 278.01, 278.02, V77.8, V85.21-V85.4) or (2) a recorded BMI that met the EHR diagnostic criteria. Current evidence indicates that, at lower BMI values, Asian patients have higher risk factors for cardiovascular diseases19 and higher rates of hypertension, dyslipidemia, and diabetes.20 As a result, the WHO has recommended that a lower BMI threshold21 be used for defining overweight in Asians.19,20 Furthermore, Asians account for 25% of the Group’s patient population, second only to non-Hispanic whites (NHW). Therefore, we used 23 for overweight for Asian patients and 25 for NHW and other races.19 Though African Americans22 and Latino23 may also have different BMI thresholds, we did not use separate thresholds for them due to their small numbers in the study population (1.4% of the sample were African American, 9.2% were Hispanic/Latino).
Hypertension was identified by: (1) ICD-9 diagnoses codes (401.x, 403.00, 403.01, 403.10, 403.11, 403.90, 403.91, 405.09, 405.11, 405.19, 405.91, 405.99), (2) prescription of antihypertensive medication with an associated diagnosis of hypertension, or (3) 2 or more high BP readings on different days within a 12-month period.21 A high BP reading was defined as>130/80 mm Hg for patients with a diabetes diagnosis either on their problem list or encounter diagnosis and ≥140/90 for patients without a diabetes diagnosis.24 Patients were considered to have hypertension during a particular year if they met the BP criteria for hypertension at any time during the year, or if they met the diagnosis or medication criteria during the year or in a previous year.
Depression was identified by: (1) ICD-9 diagnosis codes (311, 300.4, 296.2, 296.3), or (2) prescription of antianxiety, anticonvulsant, or antipsychotic medication when associated with a diagnosis of depression, or prescription of antidepressant medication.25 The list of psychiatric medications that we used to determine whether a patient was treated for depression is available in the Appendix (Supplemental Digital Content 1, http://links.lww.com/MLR/A614).
The main analyses used 2 sets of dependent variables. For high BMI analyses, these were: (1) having a diagnosis indicating overweight/obesity recorded in the EHR; and (2) the deviation of the actual BMI from the overweight threshold. For example, if a NHW person’s BMI was 33, the value of the deviation of actual BMI from the overweight threshold would be 825–33; if the patient were Asian, the deviation would be 10.23–33 The BMI values of pregnant patients were set to missing for the duration of the pregnancy and for 6 months after the end of the pregnancy.26 The hypertension-related dependent variables were: (1) having a hypertension diagnosis recorded in the EHR; and (2) systolic BP (SBP).
The explanatory variables included patient-level and primary care provider (PCP)-level variables. Time-invariant patient characteristics were: sex, age category in 2010 (18–40, 41–64 and ≥65), race (NHW, Asian, others), and initial comorbidity status. Similar to the census, race was based on patient’s self-report collected when an individual becomes a patient at the Group and entered into the EHR. To address the 38.4% of patients missing race data in the EHR, we imputed their race using a previously validated algorithm that uses both the patient’s first and last names.27 For the analysis of patients with high BMI, we used 4 variables to indicate the initial comorbidity status: (1) high BMI only (O); (2) high BMI and depression (OD); (3) high BMI and hypertension (OH); and (4) high BMI and hypertension and depression (OHD). For the analysis of patients with hypertension, we used 4 variables to indicate the starting comorbidity status: (1) hypertension only (H); (2) hypertension and depression (HD); (3) hypertension and high BMI (HO); and (4) hypertension and high BMI and depression (HOD).
Time-varying patient characteristics, measured at the patient-year level, included: comorbid disease burden (measured with the Charlson Comorbidity Index28 with scores regrouped into 0, 1, ≥2) and variables for insurance type: (1) fee-for-service, including Preferred Provider Organization and Medicare fee-for-service; (2) health maintenance organization, either commercial or Medicare health maintenance organization; and (3) other, including Medicaid, self-pay, and other. Treatment received for hypertension during the year was defined as having a recorded prescription of any antihypertensive medications during the year with a diagnosis of hypertension at any time point in the EHR. In the model for BMI trends over time, we excluded patients with cancer (other than nonmelanoma skin cancer) and patients with specific weight-changing conditions, including hyperthyroidism, polycystic ovary syndrome, eating disorders, AIDS, malnutrition, and Cushing syndrome.29 Among the remaining patients, we included an indicator variable for the presence of other weight-changing conditions (hypothyroidism) or treatment (prescription of oral corticosteroids or cyproheptadine).29
We identified the patient’s assigned PCP in the corresponding year from the EHR and included that PCP’s characteristics in the analysis. PCP sex was acquired from the Group’s website and when not available, from the License Lookup System of the Medical Board of California.30 PCP race (NHW, Asian, others) was collected from the License Lookup System of the Medical Board of California,30 with missing values inferred from PCP names and publicly available photographs. Information on PCP specialty (internal medicine or family practice) was obtained from administrative data.
Univariate analyses provide information on the study sample. Bivariate analyses summarized the relationship between the dependent variables and potential explanatory variables. Analysis of patients with high BMI was limited to patients with at least 2 BMI measurements who did not have a diagnosis of overweight/obesity in the 6 months within their first visit at the Group, and to the time points where patients had high BMI values. The analysis of the predictors of receiving a diagnosis of hypertension used the same inclusion criteria as the analysis of change in BP over time, but was further limited to the years in which the patient had high BP, as defined by at least 2 high BP readings on different days within a 12-month period.
To account for clustering of multiple observations over time at the patient level, we used multilevel models. Treatment for specific conditions, such as hypertension, is intended to affect the trajectories of those conditions, and may affect other measures. Data in the EHR on treatments for high BMI were incomplete, as some patients attended Weight Watchers or other outside services. Therefore, we did not examine treatments for patients with high BMI. The 9-item Patient Health Questionnaire,31 a depression scale, is infrequently used so we were unable to measure the probabilities of having a diagnosis for depression and for receiving treatment among those with depression.
Equation (1) illustrates a random-intercept logistic regression model for the impact of co-occurring chronic diseases during that year (Cit) on having a recorded diagnosis for overweight/obesity, or for hypertension (Diagnosisit) among patients with high BMI or hypertension, respectively. Other studies have also used the random-intercept model in longitudinal studies that include time-varying covariates.32Tit is a linear term for time, defined as the number of years (to the nearest day) since the patient’s first visit in the observation window between 2003 and 2010. In the hypertension multilevel models, this was defined as the number of years (to the nearest day) since the patient became an incident hypertensive case. Covariates included time-varying (X1) and time-invariant (X2) variables. The unit of analysis is the patient-day.
Equation (2) illustrates a multilevel model examining the impact of co-occurring chronic diseases on BMI and BP among patients with high BMI or hypertension, respectively. The model includes a random intercept and a random slope for time.where, Yit is BMIit or BPit, with BMI of individual patient i at time t, or SBP of individual patient i at time t. The coefficients of the interaction term of initial comorbidity (Ci) and time (Tit) reflect the effects of initial comorbidities on the dependent variables over time. The unit of analysis is the patient-day. The dataset was unbalanced, that is, patients could differ both in the number and time-spacing of measurements. Fortunately, the multilevel modeling approach we used does not require all patients to have the same number or time-spacing of measurements.33
Propensity Score Stratification (PSS)
We accounted for selection bias that is common to observational studies, using a longitudinal implementation of the propensity score stratification method.34 It involved estimating a random-intercept logistic regression model to account for the time-varying propensity of receiving hypertension treatment. The unit of analysis is the patient-year. The PSS model and subsequent multilevel analysis use a 6-month “clean window” (ie, absence of use) for prescription of antihypertensive medications, and a 12-month observation period after antihypertensive treatment. The clean window for prescription of medications was applied to reduce biases in an observational study arising from prevalent users. This is also known as the “new-user” design that is often used in similar observational studies.35 Explanatory variables were selected based on prior literature,36,37 and included patient demographics, prior treatment and diagnosis of hypertension, prior comorbidities, prior physician and clinic characteristics. Observations were assigned to propensity quintiles after estimation of the propensity scores. We determined that treatments were well represented in each quintile.38
Missing data on comorbidities occurred for 4% of observations. We used multilevel multiple imputations to replace the missing values.39 We eliminated 0.39% of the sample that was missing insurance data.
High BMI Analyses
Table 1 displays univariate and bivariate statistics in the first 2 sets of columns. In 2010, 18.7% of patients with high BMI had an overweight/obesity diagnosis recorded in the EHR. About 54.9% of patients had high BMI only, 10.8% had high BMI and depression, 25.9% had high BMI and hypertension, and 8.4% had all 3 conditions.
Logistic Regression Results
Factors associated with receiving an overweight/obesity diagnosis are shown in the last column of Table 1. Higher BMI was associated with higher odds of receiving a diagnosis (OR=1.6, P<0.01). Considering the significant odds ratio (3.21, P<0.01) on time (years under care at the Group) and the significant odds ratio on the interaction term of time and depression (0.84, P<0.01), they suggest that while the odds that a high BMI only patient would receive a diagnosis of overweight/obesity increased the longer she or he was cared for in the Group, this time effect was attenuated for patients with high BMI and depression compared with patients with high BMI only.
Multilevel Model of the Influence of Comorbidities on BMI Trajectory
Table 2 presents the results of the multilevel model showing the influence of initial comorbidity status on person-specific BMI trajectories. Compared with having high BMI only, having high BMI and depression at the start of the observation period was positively associated (coefficient=0.06, P<0.01) with the slope of BMI, but having high BMI and hypertension (coefficient=−0.07, P<0.01) or all 3 conditions was negatively associated (coefficient=−0.05, P<0.01) with the slope of BMI. Figure 1 shows fitted results from the model for a hypothetical patient, using characteristics of the most populous groups of patients. The simulations displayed different trajectories of BMI over time resulting from different comorbidity profiles.
Table 3 shows that, overall, in 2010, 51.7% of patients with measured high BP had a recorded hypertension diagnosis. Bivariate results suggest differences in rates of hypertension diagnosis among patients with high BP associated with factors including: whether patient’s BP was under control, number of years under care at the Group, patient age, patient race, Charlson index score, and PCP’s specialty.
Logistic Regression Results
The random-intercept logistic regression examining the probability of having a diagnosis of hypertension revealed a statistically significant effect of time (OR=3.83, P<0.01). The interaction between time and having all 3 conditions is 1.1 times the effect of time for those with BP only (P<0.01). Furthermore, patients with a Charlson score of 1 were more likely to receive a diagnosis than were patients scored 0 on the Charlson index (OR=1.8, P<0.01). The odds ratio for a Charlson score of ≥2 or more was 1.4 (P<0.05).
Multilevel Model of Influence of Comorbidity on BP Growth
None of the interaction terms between time and comorbidity statuses were significantly associated with the slope of SBP over time (Table 4). Patients with a Charlson Score ≥2 in the quintile that was most likely to be treated for hypertension had a negative coefficient for the interaction term between Charlson score and time (coefficient=−0.64, P<0.01). Those with the lowest propensity to receive hypertension treatment had a positive coefficient for the same interaction term (0.31, P<0.05).
To our knowledge, this is one of the first studies that take a longitudinal approach to analyze the impact of comorbidities on disease trajectories among American adults of all ages. The findings expand the literature on underdiagnosis of problems37,40–42 and disease trajectories.43,44
In assessing the impact of comorbidities over time, we found that the increasing likelihood of receiving overweight/obesity diagnosis associated with time was attenuated for patients with high BMI and depression, compared with those with high BMI only. This is a puzzling finding as it seemed to suggest that, over time, primary care physicians do not record overweight/obesity diagnoses for high BMI patients with depression as readily as they do with high BMI only patients. Furthermore, while co-occurring depression was positively associated with BMI growth over time, co-occurring hypertension, however, appeared to be negatively associated with BMI growth. Hypertension—or more likely, its active treatment and engagement with clinicians—seems to dampen BMI growth. In contrast, depression, or the lack of its active treatment and engagement with clinicians, seems to reduce BMI growth over time. In light of previous studies that documented the linkages between diagnoses and education,45 counseling,46 and treatments,47,48 the low rate of documentation for overweight/obesity calls for action to improve recognition of overweight/obesity.
An important implication of our findings relates to leveraging EHR for clinical decision support. It is well documented that busy clinicians are subject to cognitive biases and can overlook even obvious signs.49 For example, when a person with very high BMI did not receive an obesity diagnosis because she sought care for knee pain rather than obesity, an EHR-based clinical decision support system can automatically identify her BMI and alert her clinicians. EHR can also be useful in recognizing particular co-occurring conditions such as depression that could exacerbate overweight/obesity. A similar case can be made for patients with multiple high BP measures yet are undiagnosed for hypertension.
Our results showed several specialty-based differences. Family medicine patients who were overweight/obese based on their BMI readings had a lower probability of having diagnosis of overweight/obese recorded. In addition, family medicine patients with hypertension tended to have higher BP. Additional efforts are needed to understand whether such departmental differences are attributable to patient factors (such as the choice of an internist vs. a family physician), different care management plans, or disciplinary differences in treatment philosophies and approaches.
This study has several limitations. The data are from a single, large multispecialty group practice, most of whose patients have health insurance. We capture and control for a variety of patient characteristics to allow others to assess the applicability of the findings to other settings. Propensity score analysis adjusts for the observable factors included in the propensity model, but unlike a well-designed randomized trial, it cannot rule out unobservable factors. The physician and practice-based differences, however, should be considered hypothesis-generating. EHR data on entries of diagnoses may not accurately reflect clinical judgments made by physicians. Furthermore, treatment for high BMI was limited to data recorded in the EHR and did not capture services received outside the Group. Future work should address these limitations to better support providers in coordinating and managing care for this population, and to assist in tracking progress in improving health for individuals with MCC.
In summary, this study makes 3 contributions. It responds to a national call for epidemiologic studies of the impact of different types of comorbidities on disease trajectories. It also demonstrates a meaningful use of EHR data, which is still relatively uncommon in the literature. And, it provides a more comprehensive characterization of patients with MCC, how these conditions are documented and the treatments they have received, than is possible from an assessment of administrative data alone.
The study reveals gaps in health services among these complex patients, starting with suboptimal documentation of diseases that are present and suboptimal clinical outcomes. In light of these findings and assessments made by its own Quality Improvement efforts, the Group has made hypertension control an organization-wide priority and has implemented a team care approach that has improved BP control in several of its patient populations.
The authors thank Naihua Duan, PhD, Professor Emeritus, Department of Psychiatry, Columbia University, for advice on methods
1. Tinetti M, Fried T, Boyd C .Designing Health Care for the Most Common Chronic Condition—Multimorbidity. 2012; 307:JAMA:JAMA; 2493–2494.
3. . US Department of Health and Human Services .Multiple Chronic Conditions—A Strategic Framework: Optimum Health and Quality of Life for Individuals with Multiple Chronic Conditions. 2010 .Washington, DC:US Department of Health and Human Services.
4. Vogeli C, Shields AE, Lee TA, et al .Multiple chronic conditions: prevalence, health consequences, and implications for quality, care management, and costs.J Gen Intern Med. 2007; 22:391–395.
5. Flegal K, Graubard B, Williamson D, et al .Excess deaths associated with underweight, overweight, and obesity.JAMA. 2005; 293:1861–1867.
6. Allison DB, Fontaine KR, Manson JE, et al .Annual deaths attributable to obesity in the United States.JAMA. 1999; 282:1530–1538.
7. Groessl E, Kaplan R, Barrett-Connor E, et al .Body mass index and quality of well-being in a community of older adults.Am J Prev Med. 2004; 26:126–129.
8. Alley DE, Chang VW .The changing relationship of obesity and disability, 1988-2004.JAMA. 2007; 298:2020–2027.
9. Finkelstein EA, Fiebelkorn IC, Wang G .National medical spending attributable to overweight and obesity: how much and who’s paying? Health Aff. 2003; 22:W3-219–226.
10. Gaziano JM .Fifth phase of the epidemiologic transition: the age of obesity and inactivity.JAMA. 2010; 203:275–276.
11. Must A, Spadano J, Coakley EH, et al .The disease burden associated with overweight and obesity.JAMA. 1999; 282:1523–1529.
12. Thorpe KE, Florence CS, Howard DH, et al .Trends: the impact of obesity on rising medical spending.Health Aff. 2004; W4:480–486.
13. Blaine B .Does depression cause obesity?: a meta-analysis of longitudinal studies of depression and weight control.J Health Psychol. 2008; 13:1190–1197.
14. Roberts R, Strawbridge W, Deleger S, et al .Are the fat more jolly? Ann Behav Med. 2002; 24:169–180.
15. DiPietro L, Anda R, DW DF, et al .Depressive symptoms and weight change in a national cohort of adults.Int J Obes. 1992; 16:745–753.
16. Brady K .Weight gain associated with psychotropic drugs.South Med J. 1989; 82:611–617.
17. Needham BL, Epel ES, Adler NE, et al .Trajectories of change in obesity and symptoms of depression: the CARDIA study.Am J Public Health. 2010; 100:1040–1046.
18. Masand PS .Weight gain associated with psychotropic drugs.Exp Opin Pharmacother. 2000; 1:377–389.
19. Choo V .WHO reassesses appropriate body-mass index for Asian populations.Lancet. 2002; 360:235
20. Wildman RP, Gu D, Reynolds K, et al .Appropriate body mass index and waist circumference cutoffs for categorization of overweight and central adiposity among Chinese adults.Am Soc Clin Nutr. 2004; 80:1129–1136.
21. Katon WJ, Lin EHB, Von Korff M, et al .Collaborative care for patients with depression and chronic illnesses.N Engl J Med. 2010; 363:2611–2620.
22. Katzmarzyk P, Bray G, Greenway F, et al .Ethnic-specific BMI and waist circumference thresholds.Obesity (Silver Spring, MD). 2011; 19:1272–1278.
23. Rahman M, Berenson AB .Accuracy of current body mass index obesity classification for white, black and Hispanic reproductive-age women.Obstet Gynecol. 2010; 115:982–988.
24. Chobanian AV, Bakris GL, Black HR, et al .The Seventh Report of the Joint National Committee on Prevention, Detection, Evaluation, and Treatment of High Blood Pressure: the JNC 7 Report.JAMA. 2003; 289:2560–2572.
25. Goldman HH, Frank RG, Burnam MA, et al .Behavioral health insurance parity for federal employees.N Engl J Med. 2006; 354:1378–1386.
26. Zhao B, Wong EC, Palaniappan L .
Clinical data integration for observational studies using electronic health records: examples with defining pregnancy and gestational diabetes. Western Users of SAS Software. 2011
27. Wong EC, Palaniappan LP, Lauderdale DS .Using name lists to infer Asian Racial/Ethnic subgroups in the healthcare setting.Med Care. 2010; 48:540–546.
28. Quan H, Sundararajan V, Halfon P, et al .Coding algorithms for defining comorbidities in ICD-9-CM and ICD-10 administrative data.Med Care. 2005; 43:1130–1139.
31. Kroenke K, Spitzer RL, Williams JB .The PHQ-9: validity of a brief depression severity measure.J Gen Int Med. 2001; 16:606–613.
32. Butterworth P, Leach LS, Strazdins L, et al .The psychosocial quality of work determines whether employment has benefits for mental health: results from a longitudinal national household panel survey.Occup Environ Med. 2011; 68:806–812.
33. Singer JD, Willett JB .Applied Longitudinal Data Analysis. 2003 .New York:Oxford University Press.
34. Leon AC, Demirtas H, Li C, et al .Two propensity score-based strategies for a three–decade observational study: investigating psychotropic medications and suicide risk.Stat Med. 2012; 31:3255–3260.
35. Ray WA .Evaluating medication effects outside of clinical trials: new-user designs.Am J Epidemiol. 2003; 158:915–920.
36. Selby JV, Lee J, Swain BE, et al .Trends in time to confirmation and recognition of new-onset hypertension, 2002-2006.Hypertension. 2010; 56:605–611.
37. Banerjee D, Chung S, Wong EC, et al .Underdiagnosis of hypertension using electronic health records.Am J Hypertens. 2011; 25:97–102.
38. Faries D, Leon AC, Haro JM, et al .
Analysis of Observational Health Care Data Using SAS. Cary, NC: SAS Institute Inc.; 2010
39. Carpenter J, Goldstein H, Kenward M .REALCOM-IMPUTE software for multilevel multiple imputation with mixed response types.J Stat Softw. 2011; 45:1–14.
40. . American Geriatrics Society Expert Panel on the Care of Older Adults with Multimorbidity .Patient-centered care for older adults with multiple chronic conditions: a stepwise approach from the American Geriatrics Society.J Am Geriatr Soc. 2012; 60:1957–1968.
41. . American Geriatrics Society Expert Panel on the Care of Older Adults with Multimorbidity .Guiding principles for the care of older adults with multimorbidity: an approach for clinicians.J Am Geriatr Soc. 2012; 60:E1–E25.
42. Sorace J, Millman M, Bounds M, et al .Temporal variation in patterns of comorbidities in the Medicare population.Popul Health Manag. 2013; 16:120–124.
43. Quiñones AR, Liang J, Bennett JM, et al .How does the trajectory of multimorbidity vary across black, white, and Mexican Americans in middle and old age? J Gerontol B Psychol Sci Soc Sci. 2011; 66B:739–749.
44. Kahng SK, Dunkle RE, Jackson JS .The relationship between the trajectory of body mass index and health trajectory among older adults multilevel modeling analyses.Res Aging. 2004; 26:31–61.
45. Noel PH, Copeland LA, Pugh MJ, et al .Obesity diagnosis and care practices in the Veterans Health Administration.J Gen Intern Med. 2010; 25:510–516.
46. McAlpine DD, Wilson AR .Trends in obesity-related counseling in primary care: 1995-2004.Med Care. 2007; 45:322–329.
47. Bardia A, Holtan SG, Slezak JM, et al .Diagnosis of obesity by primary care physicians and impact on obesity management.Mayo Clin Proc. 2007; 82:927–932.
48. Waring ME, Roberts MB, Parker DR, et al .Documentation and management of overweight and obesity in primary care.J Am Board Fam Med. 2009; 22:544–552.
49. Groopman J .How Doctors Think. 2007 .New York:Houghton Mifflin Company.