Powered by the Evidence-based Practice Centers
Evidence Reports All of EHC
Evidence Reports All of EHC

SHARE:

FacebookTwitterFacebookPrintShare

Prompt Reduction in Use of Medications for Comorbid Conditions After Bariatric Surgery

Research Report ARCHIVED Jul 30, 2010
Download PDF files for this report here.

Page Contents

Note: This report is greater than 5 years old. Findings may be used for research purposes but should not be considered current.

Author Affiliations

Jodi B. Segal, M.D., M.P.H.a
Jeanne M. Clark, M.D., M.P.H.a
Andrew D. Shore, Ph.D.a
Francesca Dominici, Ph.D.a
Thomas Magnuson, M.D.a
Thomas M. Richards, Ph.D.a
Jonathan P. Weiner, Ph.D.a
Eric B. Bass, M.D., M.P.H.a
Albert W. Wu, M.D., M.P.H.a
Martin A. Makary, M.D., M.P.H.a

aJohns Hopkins University School of Medicine

Abstract

Background. Bariatric surgery leads to weight loss, but it is unclear whether surgery reduces conditions associated with obesity. We explored this by assessing the change in use of medications to treat diabetes mellitus, hypertension, and hyperlipidemia in the year following surgery.

Methods. Cohort study using administrative data from 2002–2005 from 7 Blue Cross/Blue Shield Plans. We compared the mean number of medications at the time of surgery and in the subsequent year. Medication usage by surgical patients was also compared to usage by matched-enrollees without surgery but with a propensity score suggesting obesity. With Poisson and logistic regression, we tested for statistical differences in usage accounting for repeated measures, controlling for age, sex and diabetes. We also evaluated medications expected to be less influenced by surgery (antidepressants, thyroid replacement, and antihistamines).

Results. Our cohort included 6,235 enrollees with bariatric surgery. Their mean age was 44 years with 82% women; 34% had diabetes. Medication use declined significantly by 3 months. By 12 months after surgery, medication use for diabetes, hypertension, and hyperlipidemia had declined 76%, 51%, and 59%, respectively. In contrast, thyroid hormone, antihistamine, and antidepressant use decreased by only 6%, 15% and 9%, respectively. Enrollees without surgery had a modest increase in medications for diabetes, hypertension, and hyperlipidemia of 4%, 8% and 20%, respectively.

Conclusions. Medication use for 3 serious, obesity-associated conditions decreased promptly following surgery. The clinical and economic benefits of reduced medication requirements should be considered when making decisions about the effects of bariatric surgery.

Introduction

Roughly one third of the United States population is obese.1 The prevalence is rising rapidly both in the United States and world wide.2 This will increasingly impact the health care system as obesity-related comorbidity is associated with high utilization of resources, including inpatient and outpatient services, and medications.3-6

The efficacy of bariatric surgery for weight reduction is well established.2,7 It is unclear, however, how this translates into a reduction in obesity-related complications and associated health care utilization. While several series2,7 and one randomized trial8 have demonstrated that bariatric surgery reduces the use of medications for treating diabetes, it is less clear whether surgery reduces utilization of medications for other chronic conditions that often afflict individuals with obesity and diabetes.9

To improve understanding of the effects of bariatric surgery on the highly prevalent, life-threatening comorbid conditions associated with obesity, we studied the use of medications to treat diabetes mellitus, hypertension, and hyperlipidemia and in the year following bariatric surgery. We hypothesized that use of these medications would decrease after surgery. In contrast, we hypothesized that use of medications for depression, hypothyroidism and allergic rhinitis would remain stable.

Methods

This is a historical cohort study using administrative data from January 1, 2002 through December 31, 2005.

Data Acquisition

We accessed claims data from seven Blue Cross Blue Shield health plans providing coverage in Western Pennsylvania, Philadelphia, South Dakota, Iowa, Hawaii, Michigan, North Carolina, and Tennessee. The data were de-identified in accordance with the Health Insurance Portability and Accountability Act’s (HIPAA) definition of a limited data set. The data were used in accordance with federal standards for protecting the confidentiality of the personal health information of the enrollee. The Johns Hopkins Institutional Review Board deemed the study to be exempt from Federal regulations because the research activities were considered to be of minimal risk to subjects.

We requested claims on members who met any of these inclusion criteria during 2002 through 2005: (1) completed a health risk assessment with member height and weight; (2) had a claim with a diagnosis of obesity; (3) had a paid or denied claim for bariatric surgery (see Appendix A); (4) had a paid or denied claim for a medication for promoting weight loss; or (5) were greater than 12 years old and had a diagnosis of hyperlipidemia, type 2 diabetes mellitus, sleep apnea, gall bladder disease or surgery, or metabolic syndrome. These diagnoses were identified in the claims by Common Procedural Terminology (CPT) codes, International Classification of Disease-9 Clinical Modification (ICD-9-CM) codes, National Drug Codes (NDC) or Diagnosis Related Group (DRG) codes.

The following data were acquired: (1) enrollment files for administrative data; (2) benefits information to determine medical and pharmacy coverage; and (3) inpatient, outpatient, and pharmacy claims records containing ICD-9 diagnosis, CPT codes, prescription NDC codes, and costs and charges (submitted, allowed, and paid).

Defining the Cohort

For inclusion in our analytic cohort, we required that the enrollee:

  • Have 6 months of continuous coverage, including pharmacy coverage, prior to the index date, defined as the date of bariatric surgery for patients who underwent surgery and, for those who did not have surgery, as the date of plan enrollment plus the mean time from the later of plan enrollment or January 2002 to surgery of the surgical patients (16 months)
  • Have 12 months of continuous coverage, including pharmacy coverage, after the index date
  • Be between 18 years and 64 years, inclusive
  • Not have a diagnosis of cancer of the esophagus (ICD-9 150-150.9), stomach (ICD-9 151-151.9), small intestine (ICD-9 152-152.9) or pancreas (ICD-9 157-157.9), or other digestive malignancy (DRG 172-173)

Creating Variables

Medications were grouped using the Johns Hopkins University ACG Case-Mix System (version 8.0 beta) into therapeutic categories using the NDC codes in the claims.10 It was assumed that a patient would not be on more than one drug from any therapeutic class. (Appendix B) For select outcomes, we stratified results by diagnosis of diabetes defined as at least one ICD-9 code of 250.xx or a pharmacy claim for any drug for treating diabetes prior to the index date.

We tabulated the number of unique therapeutic classes of drug that the patient had “on hand” at the time points of interest. We knew the date that each prescription was filled and the number of days of drug supplied. If this supply would result in the patient having sufficient drug “on hand” to overlap with a seven day window surrounding the time point of interest, the patient was considered to have been on this medication on that date. Medication that might be considered to be “extra” due to an early refill was added to the end of the next prescription fill for the tabulation of drug “on hand.”

We created additional variables for description and to use in modeling outcomes, including demographics (age, sex), utilization variables (hospitalizations, outpatient visits, medical and pharmacy payments) and other indicators of health status including the Resource Utilization Band from the Johns Hopkins University ACG Case-Mix System.10

Statistical Methods

We explored medication use in two ways: (1) a pre/post analysis of medication usage for surgical patients only; and (2) a pre/post analysis of medication usage for surgical patients compared to a group of enrollees predicted to be obese who did not undergo bariatric surgery. We also examined use of medications for depression, hypothyroidism, and allergies; conditions which we hypothesized would not change after surgery. We tabulated the mean number of medications at each time period for the figures.

Pre-post analysis. The primary comparisons were medication usage at 3, 6, and 12 months after surgery in comparison to usage at surgery. Medication usage was explored as the proportion of patients who were using at least one medication within a category at these times, and as the mean number of medications used per person within a category. For medication categories with only one drug (e.g., levothyroxine), these are equivalent.

For estimating the percent change in use of diabetes and antihypertensive medications post- versus pre-surgery, we used Poisson regression and general estimating equations (GEE), since we had counts ranging from 0 to 4 medications per person and repeat measures for the same enrollee.11 For the other medication classes, we used logistic regression with GEE as patients were generally on either one or no medications. A diagnosis of diabetes was included in the models as a covariate to test for differences by diabetes diagnosis where the analyses were not stratified by diabetes. Age and sex and their interactions with time were included in the models as categorical variables although they were not predictive of outcomes.

Comparison group without surgery. We identified enrollees who were predicted to be obese, but who had not undergone bariatric surgery. A model for the propensity to have a body mass index greater than 35 kg/m2 was previously developed using this administrative database and the body mass index as reported by a subset of the enrollees in their Health Risk Assessment. (Appendix C) We reviewed the distribution of the propensity scores and identified all non-surgical enrollees who had propensity scores above the 90th percentile (16% of all enrollees), while excluding women who were pregnant within one year of the index date. In the validation subset, a score above the 90th percentile had a positive predictive value for obesity of 78%. We individually matched enrollees in this upper decile from the non-surgical group to enrollees in the surgical group by exact age, sex and presence or absence of diabetes at a ratio of 3:1, if sufficient matches were available.

We tabulated medication usage for the non-surgical patients at 3-month intervals beginning at 6 months prior to their index date until one year after their index date. We estimated the change in medication use post- versus pre-surgery for the surgical patients and compared this to the change for the non-surgical group. As described above, we used Poisson and logistic regressions with GEE, while adjusting for age and sex, their interactions with time, and diagnosis of diabetes. The analyses for each class of medication were stratified by diagnosis of diabetes when the diagnosis of diabetes was significantly associated with the change in medication use over time. Analyses were done using SAS version 9.13 (SAS Institute, Cary, NC).

Results

Our cohort included 6,235 bariatric surgery patients, with 34% coded to have diabetes mellitus. Table 1 shows the characteristics of the surgical group and the comparison group of enrollees predicted to be obese. The median age of the surgical patients was 44 years, and the majority was female. Hypertension was the most prevalent comorbidity, affecting 53% of the surgical patients. The matching procedure, using the obesity propensity score, was successful as the characteristics of the patients in the two groups, before the index date, were largely similar. Many of the differences were statistically significant in this large cohort, although the clinical relevance was thought to be modest. On the whole, the surgical group had more comorbid illness than the comparison group, while the non-surgical group was older.

Change in Medication Use in the Surgical Cohort

There was a prompt decrease in the mean number of medications used in the post-surgery period for the medication classes that we hypothesized would be affected by bariatric surgery or by weight loss (Table 2). By 3 months after surgery, the mean number of medications on hand for diabetes, among enrollees with diabetes, decreased by 55%. Anti-hypertensive medication use decreased by 34% among enrollees with diabetes and by 59% among those without diabetes. Lipid-lowering therapies decreased by 55% and 52% among enrollees with and without diabetes, respectively. There was a more modest decrease in the mean number of prescriptions filled for antidepressant medications, with a 9% decrease by 12 months. Over the same time period, antihistamine use decreased by 15%. Use of thyroid replacement medication remained relatively constant.

In the analyses stratified by diagnosis of diabetes, for the classes of antihypertensive and lipid-lowering medications (Table 2), we found that patients with diabetes were on more medications at baseline than patients without diabetes. The patients with diabetes had a smaller percentage decrease in the number of antihypertensive medication at each time point after surgery than patients without diabetes, but a comparable decrease in the use of lipid-lowering medications.

Change in Medication Use Relative to Comparison Population

As shown in Table 3, medication usage prior to the index date was very similar in the surgical patients and the non-surgical comparison group. Medication usage in the year after the index date, however, differed markedly between the surgical and non-surgical groups (Figure 1 and Table 3). While there was a prompt decrease of 74% by 12 months in the mean number of diabetes medications filled by surgical patients, in the comparison groups, the number of medications filled increased by 4%. Similarly, the mean prescription fills for antihypertensive medications and lipid-lowering medications decreased markedly in the surgical group, and increased in the non-surgical group, both for patients with and without diabetes. The differences between groups were much less pronounced for antidepressant, thyroid hormone, and antihistamine medications.

Comment

The use of medications for diabetes, hypertension, and hyperlipidemia; conditions that should be responsive to weight loss or to the metabolic consequences of the surgery, decreased markedly soon after surgery. There were smaller decreases in the use of medications that we did not expect to be responsive to the surgery. Patients without diabetes had a greater reduction in their use of antihypertensive medications than patients with diabetes, which could be due to physicians’ more aggressive treatment of hypertension in patients with diabetes for nephroprotection or for prevention of cardiovascular complications.12

Our use of a comparison group strengthens the evidence that the changes observed were causally related to the surgery rather than to secular changes, or due to how pharmacy claims were recorded or our method for counting medications. When we compared the patients who had surgery to a matched group of enrollees who were predicted by our model to be obese but did not have bariatric surgery, we found that their baseline medication use was similar. There was a striking divergence in the curves for medication use after the index date. The surgical group had somewhat more comorbidity than the comparison group, but this should have been a conservative bias.

These changes in medication use happened quickly after surgery. The curves diverged by 3 months after surgery. The metabolic changes that occur early after bariatric surgery may be playing an important role in reducing needs for medication. Resolution of diabetes is likely not due to weight loss alone, but may be mediated by gastric hormones;13 the three most implicated being peptide YY (PPY) glucagon-like-peptide (GLP-1), and pancreatic polypeptide (PP). GLP-1, a known mediator of insulin regulation, increases immediately following bariatric surgery, which may explain the very rapid resolution of diabetes.14 The newer medications for diabetes targeting these pathways, exenatide and pramlintide, were not yet available during the years covered by this data. The resolution of diabetes may also be a consequence of the forced, substantial reduction in caloric intake due to the restrictions of the surgical procedures. Less clear is the mechanism for the rapid resolution of hyperlipidemia and hypertension, although this has been described previously.15,16

There is a small body of literature about depression and other Axis I psychiatric diagnoses in patients who undergo bariatric surgery, with estimates that roughly one quarter of surgical patients have affective disorders, and an additional 10% have eating disorders.17,18 There are few studies in the literature about changes in psychiatric diagnoses after bariatric surgery, with some demonstrating improvement in depressive symptoms and others documenting development of new depressive symptoms and an increased incidence of suicide postoperatively.19-22 We could not assess with this data whether the dosages of antidepressant medication changed with weight loss.

We cannot conclude definitively that bariatric surgery eliminates diabetes, hypertension, and hyperlipidemia. Indeed, we hope that the decreased use of these medications is due to resolution of these conditions, rather than physician and patient nonadherence to treatment recommendations. A recent study of bariatric surgery that used Medicare data found reductions in comorbid conditions after surgery comparable to what we observed.23 That study relied exclusively on ICD-9-CM codes to identify comorbid conditions. Our observations complement this study and advance their observations in that we examined other comorbid conditions thought not to change with surgery and in this way demonstrated the sensitivity of our methodology to the surgical intervention. Also, with the use of pharmacy claims, we were less likely to miss comorbid conditions that were under coded at the time of visits.

Limitations

With administrative data, we could only know diagnoses based on diagnostic or procedural codes. There may have been more complete coding of diagnoses among the enrollees in the surgical group, as the presence of comorbid illness is needed to assure coverage of the procedure among those with lower body mass indexes. Indeed, virtually all patients undergoing surgery had a diagnosis code for obesity while few of the non-surgical patients did. We expect, however, that the comparison group was indeed obese; obesity is consistently under-coded as a diagnosis.24,25 Administrative data are not adequate for describing the severity of individual conditions among enrollees; however, we used the well-validated ACG Case-Mix system for predicting the global burden of illness for an individual.

Our estimates of medication use were based on pharmacy claims. This only indicates that the prescription was filled; we cannot judge daily adherence to the medication. Additionally, we cannot definitively know the diagnosis for which the patient was taking a medication, particularly for those drugs with multiple indications (e.g., bupropion). For this analysis, however, filling the medication should be an adequate proxy for use of the medication because our primary interest is not a physiological measure, but is the change in medication use over time. These enrollees were all privately-insured patients and we cannot conclude that the same changes in medication utilization would be observed in patients with coverage from Medicaid or Medicare.

Future research should investigate whether these observed changes in medication use are sustained past 12 months. With clinical data, it should be verified that the change in medication fills actually signifies a decrease in the prevalence of these diseases. Additional research on the impact of bariatric surgery on other illnesses is needed, including on outcomes from other surgical procedures done after the bariatric procedure.

Conclusion

We conclude that bariatric surgery is effective for decreasing the use of medications for obesity-related diabetes, hypertension, and hyperlipidemia. This information can inform decisions about bariatric surgery and should be included in discussions with patients making decisions about bariatric surgery. Our results should be motivating to physicians caring for patients with these lethal, obesity-associated illnesses. It is conceivable that intervention with surgery may decrease the cardiovascular complications of diabetes, hypertension, and hyperlipidemia. Additionally, the possibility of eliminating medications and the resulting cost reductions and reduction in risks associated with medications may be highly valued by patients. We do not discount that there may be increases in use of other classes of medications.

References

  1. Hossain P, Kawar B, El Nahas M. Obesity and diabetes in the developing world—a growing challenge. N Engl J Med 2007;356(3):213-215.
  2. Buchwald H, Avidor Y, Braunwald E, et al. Bariatric surgery: a systematic review and meta-analysis. JAMA 2004;292(14):1724-1737.
  3. Chu SY, Bachman DJ, Callaghan WM, et al. Association between obesity during pregnancy and increased use of health care. N Engl J Med 2008; 358(14):1444-1453.
  4. Buescher PA, Whitmire JT, Plescia M. Relationship between body mass index and medical care expenditures for North Carolina adolescents enrolled in Medicaid in Prev Chronic Dis 2008;5(1):A04.
  5. Callahan CM, Stump TE, Stroupe KT, et al. Cost of health care for a community of older adults in an urban academic healthcare system. J Am Geriatr Soc 1998 Nov;46(11):1371-1377.
  6. Finkelstein EA, Fiebelkorn IC, Wang G. National medical spending attributable to overweight and obesity: how much, and who’s paying? Suppl Web Exclusives. 2003:W3-219-W3-226.
  7. MacDonald Jr KG, Long SD, Swanson MS, et al. The gastric bypass operation reduces the progression and mortality of non-insulin-dependent diabetes mellitus. J Gastrointest Surg 1997;1(3):213-220; discussion 220.
  8. Dixon JB, O’Brien PE, Playfair J, et al. Adjustable gastric banding and conventional therapy for type 2 diabetes: a randomized controlled trial. JAMA 2008;299(3):316-323.
  9. Pories WJ, Swanson MS, MacDonald KG, et al. Who would have thought it? An operation proves to be the most effective therapy for adult-onset diabetes mellitus. Ann Surg 1995;222(3):339-350; discussion 350-352.
  10. Multum Lexicon Database. Available at: http://www.multum.com/Lexicon.htm. Accessed February 27, 2008.
  11. Zeger SL, Liang KY. Longitudinal data analysis for discrete and continuous outcomes. Biometrics 1986;42(1):121-130.
  12. Abbott K, Basta E, and Bakris GL. Blood pressure control and nephroprotection in diabetes. J Clin Pharmacol 2004;44 (4):431-438.
  13. Rubino F, Gagner M, Gentileschi P, et al. The early effect of the Roux-en-Y gastric bypass on hormones involved in body weight regulation and glucose metabolism. Ann Surg 2004;240(2):236-242.
  14. le Roux CW, Aylwin SJ, Batterham RL, et al. Gut hormone profiles following bariatric surgery favor an anorectic state, facilitate weight loss, and improve metabolic parameters. Ann Surg 2006;243(1):108-114.
  15. Nguyen NT, Varela E, Sabio A, et al. Resolution of hyperlipidemia after laparoscopic Roux-en-Y gastric bypass. J Am Coll Surg 2006;203(1):24-29.
  16. Yan E, Ko E, Luong V, et al. Long-term changes in weight loss and obesity-related comorbidities after Roux-en-Y gastric bypass: a primary care experience. Am J Surg 2008;195(1):94-98.
  17. Mauri M, Rucci P, Calderone A, et al. Axis I and II disorders and quality of life in bariatric surgery candidates. J Clin Psychiatry 2008;e1-e7.
  18. Rosenberger PH, Henderson KE, Grilo CM. Psychiatric disorder comorbidity and association with eating disorders in bariatric surgery patients: a cross-sectional study using structured interview-based diagnosis. J Clin Psychiatry 2006;67(7):1080-1085.
  19. Nickel C, Widermann C, Harms D, et al. Patients with extreme obesity: change in mental symptoms three years after gastric banding. Int J Psychiatry Med 2005;35(2):109-122.
  20. Scholtz S, Bidlake L, Morgan J, et al. Long-term outcomes following laparoscopic adjustable gastric banding: postoperative psychological sequelae predict outcome at 5-year followup. Obes Surg 2007 Sep;17(9):1220-1225.
  21. van Hout GC, Boekestein P, Fortuin FA, et al. Psychosocial functioning following bariatric surgery. Obes Surg 2006;16(6):787-794.
  22. Nickel MK, Loew TH, and Bachler E. Change in mental symptoms in extreme obesity patients after gastric banding, Part II: Six-year follow up. Int J Psychiatry Med 2007;37(1):69-79.
  23. Perry CD, Hutter MM, Smith DB, et al. Survival and changes in comorbidities after bariatric surgery. Ann Surg 2008;247(1):21-27.
  24. Huang J, Yu H, Marin E, et al. Physicians’ weight loss counseling in two public hospital primary care clinics. Acad Med 2004;79(2):156-161.
  25. Hauner H, Koster I, von Ferber L. Frequency of ‘obesity’ in medical records and utilization of out-patient health care by ‘obese’ subjects in Germany. An analysis of health insurance data. Int J Obes Relat Metab Disord 1996;20(9):820-824.

Tables and Figure

 

Table 1. Characteristics of the bariatric surgery group and the matched comparison group of non-surgical patients likely to be obese.
  Surgical Group
(N=6,235)
Non-Surgical Comparison Group
(N=16,116)
CHF=congestive heart failure; GERD=gastroesophageal reflux disease; IHD=ischemic heart disease, NA=not applicable;
* from the ACG-case Mix System—an indicator of resource utilization
Age in years, % 18-34 20 6.8
35-44 30 35
45-54 34 40
55-65 16 19
Mean years (SE) 44 (0.12) 46 (0.07)
Male, % 18 21
Year of Enrollment, % 2002 84 84
2003 13 16
2004 3.8 0
Type of Surgery, % Gastric Bypass 80 NA
Gastric Banding 1.6 NA
Other 18 NA
Comorbid Conditions, % Hypertension 53 47
Diabetes 34 39
Hyperlipidemia 23 19
IHD 7.8 5.2
CHF 2.2 2.4
Depression 19 7.8
Sleep Apnea 30 11
GERD 36 8.5
Obesity 97 5.2
Resource Utilization Band*,
(% in each percentile category)
1 -20 0.7 8.2
21-40 3.2 17
41-60 67 60
61-80 22 11
81-100 7.3 4.2
Total medical costs in 6 months preceding surgery, $ (mean SE) 4,046 (57) 2,421 (40)
[median] [2,061] [876]
Total pharmacy costs in 6 months preceding surgery, $ (mean SE) 1,231 (20) 1,199 (13)
[median] [734] [754]

 

Table 2. Medication use per person following bariatric surgery.*
Medication Class Average Medication Count at Surgery
[95% Confidence Interval]
Count at 3 months (and % decrease) Count at 6 months (and % decrease) Count at 12 months(and % decrease)
* predicted results for females aged 45 to 54 years; all p-values for change over time ≤ 0.0001 unless specified
controlled for presence of diabetes,
p ≤ 0.05,
§ p ≤ 0.001
Diabetes Diabetic Patients Only: 1.1
[0.96-1.2]
0.45 (58) 0.35 (68) 0.27 (75)
Anti-hypertensives Diabetic Patients: 1.0
[0.91-1.1]
0.66 (34) 0.61 (39) 0.55 (45)
Non-diabetic Patients: 0.72
[0.67-0.78]
0.41 (43) 0.37 (48) 0.33 (54)
  Probability of Medication at Surgery
[95% Confidence Interval]
Probability of Medication at 3 Months (and % decrease) Probability of Medication at 6 Months (and % decrease) Probability of Medication at 12 Months (and % decrease)
Lipid-lowering Diabetic Patients: 0.34
[0.33-0.37]
0.15 (55) 0.16 (52) 0.15 (55)
Non-diabetic Patients: 0.16
[0.14-0.18]
0.076 (52) 0.065 (59) 0.065 (59)
Anti-depressants All Patients 0.39
[0.37-0.41]
0.33 (15) 0.34 (12) 0.36 § (9)
Thyroid Replacement All Patients 0.17
[0.15-0.19]
0.16 (4.1) 0.16 (4.6) 0.16 (6.6)
Antihistamines All Patients 0.10
[0.083-0.11]
0.074 (23) 0.078 (19) 0.082 (15)

 

Table 3. Medication use in the bariatric surgery group and the matched non-surgical group of patients likely to be obese.*
Medication Class Average Count of Medications at Index Date
[95% Confidence Interval]
Average Count of Medications at 12 months
[95% Confidence Interval]
(and % change from index date)
Surgical Group Non-surgical Group Surgical Group Non-surgical Group
* adjusted for age and sex and their interaction with time with results predicted for females aged 45 to 54 years
controlled for presence of diabetes
p<0.0001 for difference in change between the two groups
Diabetes Diabetic Patients 1.07
[1.02-1.12]
1.15
[1.12-1.19]
0.28
[0.25-0.31]
(74% decrease)
1.20
[1.16-1.24]
(4% increase)
Antihypertensives Diabetic Patients 1.03
[0.98-1.09]
1.06
[1.02-1.10]
0.55
[0.51-0.60]
(46% decrease)
1.10
[1.06-1.14]
(4% increase)
Non-diabetic Patients 0.69
[0.66-0.73]
0.93
[0.90-0.96]
0.31
[0.28-033]
(55% decrease)
1.02
[0.98-1.05]
(9% increase)
  Probability of Patient Using Medication at Index Date
[95% Confidence Interval]
Probability of Patient Using Medication after 12 months
[95% Confidence Interval]
(and change from index date)
Surgical Group Non-surgical Group Surgical Group Non-surgical Group
Lipid-lowering Diabetic Patients 0.34
[0.32-0.37]
0.35
[0.33-0.37]
0.16
[0.14-0.17]
(54% decrease)
0.39
[0.37-0.40]**
(10% increase)
Non-diabetic Patients 0.15
[0.14-0.17]
0.20
[0.19-0.22]
0.064
[0.06-0.07]
(59% decrease)
0.25
[0.24-0.26]**
(23% increase)
Antidepressants All Patients 0.39
[0.36-0.40]
0.26
[0.24-0.26]
0.36
[0.33-0.36]
(9% decrease)
0.27
[0.25-0.27]
(3% increase)
Thyroid Replacement All Patients 0.18
[0.16-0.18]
0.16
[0.15-0.18]
0.16
[0.15-0.18]
(6% decrease)
0.17
[0.16-0.19]
(4% increase)
Antihistamines All Patients 0.098
[0.07-0.09]
0.086
[0.08-0.09]
0.083
[0.08-0.09]
(15% decrease)
0.086
[0.08-0.09]
(1% increase)

Figure 1. Mean medication use over time in surgical patients and in a non-surgical comparison group. The figure shows six graphs. Each graph compares medication use over time in the surgical group with a nonsurgical group of individuals who were predicted to be obese. Medication use is represented as the mean number of medications used per person. Medication use is shown at 3- month intervals, starting with 6 months before the date of bariatric surgery for patients who underwent surgery and ending with 12 months after the surgery. By 3 months after surgery, the mean number of diabetes medications among enrollees with diabetes decreased to approximately 0.15 from approximately 0.35 at the time of surgery. The mean number of diabetes medications continued to decrease, and at 12 months it was approximately 0.10. In the nonsurgical group, the mean number of diabetes medications was approximately 0.35 at 3 months and it increased to about 0.5 at 12 months. Antihypertensive medications and lipid-lowering therapies also decreased in the surgical group and remained much lower than in the nonsurgical group at 12 months. For antidepressant medications, there was a more modest decrease in the mean number of medications in the surgical group. For thyroid replacement medications, use remained relatively constant in the surgical and nonsurgical groups.
White Diamonds=Surgical Group
Black Squares=Non-surgical Group of Individuals Predicted to be Obese

Appendixes

 

Appendix A. Bariatric Operations

DRG codes 288 Procedures for Obesity Other
1 Must be accompanied by DRG 288
2 Must be accompanied by DRG 288 or another bariatric surgery procedure

DRG = Diagnosis-Related Groups; CPT = Current Procedural Terminology;
HCPCS = Health Care Common Procedure Coding System, Level II;
ICD-9-CM = International Classification of Diseases revision 9, Clinical Modification

CPT codes 43644 Laparoscopy, surgical, gastric restrictive procedure; with gastric bypass and Roux-en-Y gastroenterostomy (roux limb 150 cm or less) Gastric bypass
43645 Laparoscopy, surgical, gastric restrictive procedure; with gastric bypass and small intestine reconstruction to limit absorption Gastric bypass
436592 Unlisted laparoscopy procedure, stomach Other
438101 Gastroduodenostomy Other
438201 Gastrojejunostomy without vagotomy Other
438251 Gastrojejunostomy with vagotomy any type Other
43842 Gastric restrictive procedure, without gastric bypass, for morbid obesity; vertical-banded gastroplasty Banding
43843 Gastric restrictive procedure, without gastric bypass, for morbid obesity; other than vertical-banded gastroplasty Other
43845 Gastric restrictive procedure with partial gastrectomy, pylorus-preserving duodenoileostomy and ileoileostomy (50 to 100 cm common channel) to limit absorption (biliopancreatic diversion with duodenal switch) Other
43846 Gastric restrictive procedure, with gastric bypass for morbid obesity; with short limb (150 cm or less) Roux-en-Y gastroenterostomy Gastric bypass
43847 Gastric restrictive procedure, with gastric bypass for morbid obesity; with small bowel reconstruction to limit absorption Gastric bypass
439992 Unlisted procedure, stomach Other
442382 Unlisted laparoscopy procedure,intestine (except rectum) Other
HCPCS codes S2082 Laparoscopy, surgical; gastric restrictive procedure, adjustable gastric band includes placement of subcutaneous port Banding
S2085 Laparoscopy, gastric restrictive procedure, with gastric bypass for morbid obesity, with short limb (less than 100 cm) Roux-en-Y gastroenterostomy (code no longer in use after 12-31-04) Gastric bypass
ICD-9-CM procedure codes 4351 Partial gastrectomy Other
4361 Distal gastrectomy Other
4371 Partial gastrectomy with jejunal anastomisis Other
43891 Sleeve gastrectomy Other
4431 High gastric bypass Gastric bypass
44382 Laparascopic gastroenterostomy Other
44392 Gastroenterostomy NEC Other
44682 Laparoscopic gastroplasty Other
44932 Gastric bubble insertion Other
4495 Laparoscopic gastric restrictive procedure Other
44992 Gastric operation not elsewhere classified Other
45501 Isolated intestinal bypass, small bowel to small bowel anastomosis Other
45511 Isolated intestinal bypass, small bowel to segment isolation Other
45901 Isolated intestinal bypass, intestine to intestine anastomosis not otherwise specified Other
45911 Isolated intestinal bypass, intestinal isolation not otherwise specified Other

 

Appendix B. Medications of Interest

Medication Category Therapeutic Classes
* In 11/2002, loratadine became available over the counter.
Antihypertensive Medications angiotensin converting enzyme inhibitor
calcium-channel blocker
angiotensin receptor blocker
diuretic
beta-blocker
other antihypertensive medications
Lipid-lowering Medications HMG-CoA reductase inhibitor
fibrates
niacin
bile acid sequestrant
other lipid lowering therapies
Diabetes Treatments insulin
pramlintide
sulfonylureas
biguanides
thiazolidinedione
alpha glucosidase inhibitors
meglitinides
glucagon-like peptide agonist
Antidepressants tricyclic antidepressants
selective serotonin reuptake inhibitors
other antidepressants
Thyroid Replacement not further classified
Antihistamines* not further classified

 

Appendix C. Description of the Development and Validation of a Propensity Score for Obesity

Introduction

Obesity is associated with many comorbidities and disability. Obesity is typically under-coded by practicing physicians, hampering efforts for disease management or research on obesity using administrative data.C1-C4 Our objective was to develop a propensity score model based on clinical data found in health plans claims files. The ultimate goal was to identify patients with Class II or III obesity (BMI ≥ 35 kg/m2). For this project this tool was used to identify a non-surgical cohort to serve as a comparison group for a cohort of patients undergoing bariatric surgery.

Methods

We used data from “health risk appraisal” (HRA) surveys from 3 participating BCBS plans, which included self-reported height and weight, and linked it to claims data from 2002-2005 (N=115,495). We then excluded records with any of the following:

  • < 6 months coverage in the year in which the HRA was completed (N=16,810)
  • Missing data regarding age or age <18 years (N=135)
  • Had a bariatric surgery claim during the study period (N=171)
  • Had a pregnancy claim during the study period (N=3,493)
  • BMI unable to be calculated or BMI <10 kg/m2 or >100 kg/m2 (N=625)

Our final sample (N=71,057) was randomly split in two subsamples, one for development (N=35,529), and one for validation (N=35,528).

Our dependent outcome was class II or III obesity, defined by a BMI ≥35 (from self-reported height and weight). In addition to age and gender, we used ICD-9-CM diagnosis codes that we categorized using the Expanded Diagnosis Clusters (EDC) clustering system as our predictors. We also used prescription drug claims information (NDC codes) to identify additional persons under treatment for disease who may not have been identified using ICD diagnosis codes). This system for categorizing NDC codes based on the likely condition being treated is known as the Rx Morbidity Group (RxMG) system. Both of these disease markers methodologies are part of the widely used and validated Johns Hopkins ACG case-mix / predictive risk methodology (See http://www.acg.jhsph.edu).C5,C6

We conducted bivariate logistic regression analyses to determine which covariates were associated with obesity. We then conducted multivariate logistic regression analyses in several phases using: (1) all variables, (2) stepwise regression to select variables with p<0.10, (3) variables with odds ratios >2.0 or < 0.5, and (4) variables anticipated to be associated (+ or -) based on clinical expertise. We reviewed and compared all models and selected a final model. We then tested the model in the second half of the sample. We examined the model by applying it to a large sample of enrollees in 5 participating BCBS plans using data from same time period.

Results

The comparison of the performance of different predictive (propensity) models is shown in Table C-1. We present the “C” statistic (based on “receiver operating characteristics” – ROC, also known as area under the curve).

In our model, the ICD-9-based “obesity EDC” had a very significant and sizable predictive coefficient. (That is, this code significantly contributed information to the prediction that a person’s BMI was greater than 35kg/m2). For case finding purposes, in general populations, this model would be quite useful. Every person receiving bariatric surgery had this EDC code because all persons receiving a bariatric procedure required a hospital diagnosis of obesity for payment of the claim. However, we found that only about 15% of those persons with a known BMI greater than 35 kg/m2 (but without bariatric surgery) were coded as having an obesity diagnosis by their providers. Thus the use of this diagnosis code differed among obese persons in our two study cohorts (i.e., those obese persons undergoing surgery versus those not receiving surgery). Therefore, we opted to exclude this single EDC from the final obesity propensity model.

As noted on Table C-1, the final “parsimonious” model included a selection of EDC and RxMG categories while excluding the obesity EDC code. The final model had an ROC of 0.714 in the validation sample.

Table C-2 presents the sensitivity, specificity and positive predictive value (PPV) for different levels of the propensity score for the final model, within the validation half of the HRA survey population. For those persons whose claims-based propensity score fell into the highest 5% percentile, fully 96% reported BMIs greater than 35kg/m2. This very high specificity suggests that the propensity score can effectively be applied to claims data files to identity a cohort that is extremely likely to be obese.

 

Table C-1. Comparison of the performance of different models in the validation sample (N=35,528).
Risk Model Based on Claims Data C-Statistic
* Model chosen to define comparison group
Full EDC Model (All 200+ diagnostic categories) 0.718
Parsimonious EDC model (w/o obesity diagnosis) 0.702
Full RxMG Only (NDC codes only; 50+ Rx defined disease/ condition categories) 0.674
Full EDCs + RxMGs (ICD + NDC) 0.731
Final Parsimonious Model with EDC’s and RxMG’s*
(w/o Obesity Indicator; 63 EDCs & 19 RxMGs)
0.714

 

Table C-2. Screening characteristics of the selected propensity model in the validation sample.
Percentile of Propensity Score Sensitivity Specificity PPV
Top 1% 0.06 0.99 0.92
Top 5% 0.19 0.96 0.83
Top 10% 0.29 0.92 0.78
Top 25% 0.52 0.77 0.69

 

References

C1. Hauner H, Koster I, von FL. Frequency of ‘obesity’ in medical records and utilization of out-patient health care by ‘obese’ subjects in Germany. An analysis of health insurance data. Int J Obes Relat Metab Disord 1996;20:820-824.

C2. Huang J, Yu H, Marin E, et al. Physicians’ weight loss counseling in two public hospital primary care clinics. Acad Med 2004;79:156-161.

C3. Bramlage P, Wittchen HU, Pittrow D, et al. Recognition and management of overweight and obesity in primary care in Germany. Int J Obes Relat Metab Disord 2004;28:1299-1308.

C4. O’Brien SH, Holubkov R, Reis EC. Identification, evaluation, and management of obesity in an academic primary care center. Pediatrics 2004;114:e154-e159.

C5. Starfield B, Weiner J, Mumford L, et al. Ambulatory care groups: a categorization of diagnoses for research and management. Health Serv Res 1991;26:53-74.

C6. Weiner JP, Starfield BH, Steinwachs DM, et al. Development and application of a population-oriented measure of ambulatory care case-mix. Med Care 1991;29:452-447.

Project Timeline

Effectiveness and Safety of Bariatric Surgery in Patients with Type II Diabetes Mellitus and in Patients at Risk for Gestational Diabetes: A Retrospective Cohort Study for the DEcIDE Research Network

Jul 28, 2010
Topic Initiated
Jul 30, 2010
Research Report Archived
Page last reviewed January 2020
Page originally created November 2017

Internet Citation: Research Report: Prompt Reduction in Use of Medications for Comorbid Conditions After Bariatric Surgery. Content last reviewed January 2020. Effective Health Care Program, Agency for Healthcare Research and Quality, Rockville, MD.
https://effectivehealthcare.ahrq.gov/products/bariatric-surgery/research

Select to copy citation