Use of International Classification of Diseases,Ninth Revision Codes for Obesity: Trends in the United States from an Electronic Health Record-Derived Database

Introduction

Obesity has become one of the most important public health challenges of the modern era. Globally, the prevalence of obesity has grown at an alarming rate over recent decades, and it has more than doubled since 1980. ¹ Data from the National Health and Nutrition Examination Survey, 2011–2012, showed that more than one third of adults in the United States – more than 78 million individuals – were obese, with the prevalence being higher in middle-aged than younger adults. ² The medical consequences of obesity include multiple comorbidities such as type 2 diabetes mellitus, fatty liver disease, hyperlipidemia, hypertension, coronary artery disease, obstructive sleep apnea, and osteoarthritis, among others, ^{3 –6} adding to the morbidity, mortality, and economic burden of obesity. ^{7 –9} Given that obesity is a potentially modifiable risk factor, there is an urgent need to develop a greater understanding of its relationships to other diseases and impact on health care utilization, cost, and medical outcomes. In other disease states, this often is analyzed in the context of claims data generated from International Classification of Diseases (ICD) coding of primary diseases, complications, and interventions. However, if overweight and obesity are not reliably coded in these databases, then the study of the subsequent complications and interventions will be limited and potentially biased.

In order to study the nature and scope of obesity-related complications, whether it be from a medical or payer perspective, it is critical that the population being studied is identified accurately and completely for the findings to be valid. Although medical claims data are well recognized as useful for large-scale evaluations of disease epidemiology, including medical and economic outcomes, previous research has demonstrated that coding for obesity and body mass index (BMI) in medical claims is inconsistent at best and is significantly underrepresentative of populations with overweight or obesity. ^{10 –16} The reasons are numerous, but it is likely that ICD-9 coding for overweight and obesity is often overlooked unless it is the primary reason patients are seeking treatment. In addition, there are provider reimbursement challenges for obesity management, thus making the reimbursement code alone not well suited to identifying patients who may be overweight or obese.

Increased adoption of electronic health record (EHR) technology by physicians in the United States is providing a new source of health care data for outcomes research purposes. Such data are based on information actually measured during the routine process of patient care and include biometric data, such as BMI, not available in claims databases. The Quintiles electronic medical record (EMR; QuintilesIMS, Durham, NC) research database is a commercially-available, high-quality source of anonymized patient-level ambulatory medical records. This database captures a range of demographic and clinical variables, including BMI, and is increasingly being used for outcomes research pertaining to various disease states. ^{17 –22}

The primary objective of this study was to determine the prevalence and accuracy of ICD-9 coding for obesity and overweight within the Quintiles EMR, compared to BMI values calculated from patient data. The study also was designed to analyze characteristics of patients with overweight or obesity in the database (based on BMI values) who did or did not receive ICD-9 codes for overweight or obesity. A third objective was to examine whether the use of ICD-9 coding for overweight and obesity changed between 2011 and 2014.

Methods

Results

Part II: Coding trends over time

Among patients with BMI ≥30 kg/m², the prevalence of ICD-9 coding for obesity increased slightly each year from 2011 to 2014 (Fig. 1), with a slightly greater increase apparent between 2013 and 2014 than between other years, particularly within BMI categories ≥35 kg/m². For all years, the prevalence of coding for obesity was low, but it increased as BMI category increased. Similar trends were observed for both males and females, though the prevalence of coding was consistently higher among women than men (Table 5). Similar trends were observed for coding among males and females even when analyzed by age categories (data not shown).

OPEN IN VIEWER

FIG. 1.

Prevalence of coding for obesity^a by index body mass index^b category, by year, 2011 to 2014.^c

OPEN IN VIEWER

Table 5.

Prevalence of International Classficiation of Diseases, Ninth Revision Coding for Obesity ^a by Index Body Mass Index ^b by Year and Sex

	% of patients with ICD-9 codes for obesity a
	2011		2012		2013		2014
BMI, kg/m2	Females	Males	Females	Males	Females	Males	Females	Males
25.0–29.9	2.4	1.3	2.5	1.5	2.7	1.6	3.3	2.0
30.0–34.9	12.1	9.8	13.2	10.7	15.2	12.4	17.9	14.4
35.0–39.9	23.7	23.3	24.5	23.8	26.4	25.2	30.4	28.8
40.0–44.9	34.7	36.7	35.0	36.6	36.5	37.9	41.4	42.2
45.0–49.9	43.4	47.9	43.3	46.6	44.7	48.2	49.8	52.5
≥50.0	54.5	58.9	54.2	58.1	55.1	58.7	59.9	63.4

a

ICD-9 codes 278, 278.0, 278.00, 278.01, 278.03.

b

Index BMI = first recorded BMI measurement during the study period.

BMI, body mass index; ICD-9, International Classficiation of Diseases, Ninth Revision.

Similar trends were observed for coding for overweight, though coding for overweight was observed in very low proportions of patients overall (online Supplementary Fig. S1). Among patients with a BMI consistent with the classification of overweight (BMI 25–29 kg/m²), the proportion of patients with ICD-9 coding for overweight was very low: 1.8% in 2011, 2.4% in 2012, 3.4% in 2013, and 3.9% in 2014. Prevalence of ICD-9 coding for overweight by year and sex is provided in online Supplementary Table S4.

Discussion

This study provided a unique opportunity to examine the prevalence and trends associated with clinician ICD-9 coding for overweight and obesity based on actual patient BMI data. Results demonstrate that ICD-9 codes should not be relied on to estimate population prevalence rates of obesity or overweight, given the likelihood of capturing only a portion of relevant patients and generating inaccurate conclusions about the magnitude of the obesity problem. Furthermore, even in other conditions where weight may be an effect modifier, using ICD-9 codes for obesity/overweight will not provide an effective control of confounding.

Even for patients in the very highest BMI categories of 40 kg/m² and greater only one third to just over one half were coded for obesity in any given year. Although the frequency of coding for all categories of overweight or obesity increased each year of the study, the overall frequency remained extremely low for such an important disease.

Interestingly, patients with obesity with rarely coded comorbidities such as Prader-Willi Syndrome, metabolic syndrome, sleep apnea, nonalcoholic fatty liver disease, and Cushing syndrome were the most likely to receive codes for obesity. Among this group, Prader-Willi Syndrome and Cushing Syndrome are epidemiologically uncommon, occurring very rarely in any population and are specialist treated. The fact that obesity is a core presentation of both of these conditions may increase its likelihood for coding. In contrast, the other comorbid conditions where obesity coding is common, such as metabolic syndrome, sleep apnea, and nonalcoholic fatty liver disease are actually epidemiologically quite common; however, they are also commonly underdiagnosed and, thus, likely to also be undercoded. In the case where one of these comorbidities has been diagnosed and is being treated, coding alongside obesity coding may not be surprising, as each of these comorbidities is driven by obesity. Additionally, because it is not common to code for these comorbidities or for obesity, it is possible that there are particular clinicians who are more likely to either assess these conditions and/or code for them. Surprisingly, the rate of obesity coding was lower among other obesity-driven comorbidities such as type 2 diabetes, hypertension, and dyslipidemia. However, perhaps their high population prevalence, in addition to the comorbid presentation of all 3, does not provide an automatic trigger to clinicians to also code for obesity.

Possible reasons for incomplete obesity and overweight coding might include the time burden of documentation, including prioritizing other codes in the limited time visit, shortage of time to search for the code, lack of knowledge of the existence of ICD codes for overweight and obesity, EHR system issues wherein the codes do not come up in searches, the limit of only 4 codes per procedure/visit (higher prioritization for “mainstream” diseases), and a nihilistic view of coding weight status related to a perception or reality of limited or no reimbursement for management of obesity, because there currently is no increased payment or minimal economic incentive to do so with many payers. It is also worth noting that the data found an unexpected “false positive” phenomenon in which a small percentage of patients with very low BMIs (<25 kg/m²) were coded as overweight or obese, which could suggest a certain level of operational or administrative error when it comes to coding accuracy in general.

A lack of obesity coding behavior also may reflect a general aversion among clinicians toward addressing the topic of weight with their patients. Reasons for such behavior can include time limitations, failure to prioritize obesity as a clinical issue, negative stereotypes and attitudes about people with obesity, low levels of emotional rapport with people with obesity, and poor expectations for weight loss success. ^{25 –28} Some of these challenges may be addressed by improving health care provider training in weight management.

Another likely phenomenon of the obesity epidemic is that, as the mean BMI in the US has risen, patients with BMIs >25 kg/m² (overweight) or >30 kg/m² (obese) are subconsciously assessed as reflecting the norm rather than a population at increased health risk that requires intervention. In addition, clinicians are likely unaware of the biological basis of obesity, and hence may be cynical about their ability to facilitate weight loss. This is probably driven by past disappointments with diet and exercise interventions or historic failures of older weight loss medications. Data from the National Ambulatory Medical Care Survey revealed a significant decline in primary care physician-based weight counseling efforts from 1995–1996 to 2007–2008 in the United States. ²⁹ This is an alarming trend, given the millions of individuals with overweight/obesity, which has well-established health consequences such as diabetes, heart disease, fatty liver, and others, and the fact that overweight/obesity is a modifiable risk factor.

A number of major changes have occurred in recent years with potential impact on the medical coding environment in the United States. The US Health Information Technology for Economic and Clinical Health Act, passed in 2009, provides financial incentives to hospitals and outpatient clinics that utilize EHRs meeting certain requirements, known as “meaningful use” standards. As part of these standards, BMI calculations must be included in the EHRs. Adoption of EHR compliance among outpatient providers reportedly rose from 17% in 2006 to 78% in 2013. ³⁰ The Affordable Care Act (ACA) went into full effect in January, 2014, providing millions of new patients access to the US health care system overall, though adding additional administrative burdens to medical practices as they adjust to greater patient demand, multitudes of regulations, and adapt to additional payer models. However, the ACA also has strict documentation rules, reinforcing the need for medical coding accuracy to avoid denial of reimbursement. The current data noted a slight but noticeable increase in obesity coding prevalence in 2014 relative to other years, most evident among patients with BMI ≥35. Although this finding coincides with the full implementation of the ACA in the same calendar year, possible causal associations cannot be determined. The ACA does require full coverage for preventive care obesity screening and counseling by most insurers, possibly increasing the financial incentive for physicians to code and bill for weight-related services. Further studies looking at coding trends over time will be needed to evaluate the ongoing impact of the ACA and Medicare and commercial payer quality reporting requirements on coding compliance and accuracy.

Another change affecting the US health care system, and which the current data do not reflect, is the implementation of the ICD-10 coding system in October 2015. The ICD-10 system represents a tremendous expansion of coding complexity, as it contains approximately 72,000 diagnosis codes compared with 14,000 in the ICD-9 system. Previous numeric ICD codes for diseases have been replaced with a vastly expanded set of alphanumeric codes. Although the intention is greater specificity and accuracy, it remains to be seen if this new system will achieve the desired goals given the much greater effort on behalf of the clinical team to search, identify, and document these new codes. The present data may provide a useful benchmark for future studies to examine obesity coding trends with the ICD-10 system.

Previous US studies of varying methodology and sample size have reported under-recognition and poor coding practices for obesity, including studies utilizing Medicare records, ³¹ the Mayo Clinic Rochester primary care database, ¹³ and US military health system EMRs. ¹⁴ The present data corroborate and expand on these prior findings, providing additional robust evidence from a multimillion patient sample with broad national representation of outpatient coding practices in the United States. The present findings add to a growing body of evidence substantiating the critical limitations of relying on medical coding for tracking obesity burden, epidemiology, expenditures, and outcomes.

Although not all providers will accept and enact the requirement, as of January 2014 all health care providers in the United States are now federally mandated to use EHR systems. Hopefully, this will increase the scope and utility of such databases for outcomes research, though it is unclear to what extent researchers will have access to these improved data sets.

What should be equally concerning about these data is the implication that if clinicians are not coding obesity, they likely are not treating obesity ¹³ ; prior research has revealed that documentation of obesity correlates with interventional recommendations. ³² Given that two thirds of the US population is overweight or obese, the medical community needs to proactively embrace treating overweight and obesity much as they do for other chronic diseases such as depression or hypertension. In fact, the magnitude of the problem represents a tremendous opportunity for clinicians to make an impact on numerous weight-related diseases simply by maintaining awareness of obesity as a disease and being proactive in recommending evidence-based and effective weight-loss measures. Proactive physician communication, intervention, and advice have been shown to be effective motivators for weight-loss behavior, ³³ and medical therapy is often indicated.

Conclusions

This analysis found that patients with measured BMIs indicating overweight or obese status in outpatient settings in the United States still are not reliably coded in claims data as such. Physicians, payers, and administrators who are dedicated to evaluating medical and economic outcomes related to obesity must be aware that medical coding data for overweight and obesity are wholly inadequate for these purposes. BMI determinations from clinical height and weight data available within EHR databases are, in theory, easily accessible and provide an accurate, objective means of classifying overweight and obesity status. The apparent high degree of discordant obesity coding may signal a need for efforts to correct underlying perceptions and attitudes of physicians regarding the medical relevance of obesity and the need for effective intervention.