Estimation of the Mismatch between Admission and Discharge Diagnosis for Respiratory Patients, and Implications on the Length of Stay and Hospital Charges

Abstract

Admission and discharge diagnoses in hospitals are often in discord, and this has significant implications for the cost of care and patient safety. In this paper we used medical claims data to examine these differences for beneficiaries with respiratory conditions and quantified the degree to which specific respiratory conditions are mistaken for other ones, on admission. Since respiratory problems have seasonality, we performed two separate analyses, for summer and for winter admissions. The length of stay and hospital charges were compared between matching and non-matching {admission, discharge Dx} pairs, using independent samples t-test analysis. Results were integrated into a standalone application where physicians can select an admission diagnosis to see (i) the probability for this diagnosis to be correct (matching the discharge Dx), (ii) the probabilities for mismatch and (iii) pair-specific differential diagnosis criteria to consider reassessing the patient before confirming the admission diagnosis.

Introduction

Respiratory diseases affect the lungs and other parts of the upper and lower respiratory system. Hundreds of millions of people of all ages suffer from -often preventable- chronic respiratory diseases and respiratory allergies, globally. [1] Every year more than 3 million people die only because of Chronic Obstructive Pulmonary Disease (COPD), while it is estimated that 235 million people suffered from asthma in 2017 [2]. Respiratory diseases include conditions like asthma, COPD, pneumonia, bronchitis, and other upper and lower respiratory infections. People are more susceptible to some respiratory conditions during winter, rather than in the summer [3]. This is reflected to some extend in hospitalization rates [4]. Medical claims data for Medicare patients, for instance, inform us that patients discharged from US hospitals with a primary diagnosis (Dx) of pneumonia make up the 4.6% of the total admissions in summer, versus 4.9% in winter months. In non-tropical climates, during winter, this vulnerability can be attributed to air dryness and to people being in closer quarters indoors, making it much easier to spread a contingent respiratory infection [5]. People who have already been diagnosed with such respiratory conditions are more likely to get affected by such diseases again [5].

Hospital admissions due to respiratory problems are very frequent. Newly admitted patients are assigned with an admission Dx code upon hospitalization. This code on the institutional claim indicates the beneficiary’s initial Dx at the time of admission. The principal discharge Dx is the condition that occasioned the need for hospitalization and is determined after the patient has been thoroughly examined and completed the diagnostic examinations. The admission and discharge Dx’s are often in discord. Researchers have studied the discrepancy between admission and discharge diagnoses and its effect on outcomes and other hospital attributes [6]. Even since the 70s, researchers were interested to study these discrepancies and the characteristics of patients with high rates of admission-discharge Dx discrepancies. Leske et al. found that discrepancies existed in 26.8% of all hospital admissions, and that those were most frequent in neurological, medical and pediatric patients [7]. Austin et al. studied the coding accuracy of hospital discharge data for patients admitted to cardiac care units in Canada and found that the sensitivity of the diagnoses under study was only 60.7% [8]. Discrepancies between admission and discharge Dx may lead to unwanted medical examinations, incorrect treatments, or delays to deliver the appropriate care to the patient. There are also very important patient safety implications, and unwanted hospital outcomes including an increased hospital length of stay, increased cost of care, which may be transferred to patients and their payers in the form of increased hospital charges.

The few aforementioned existing approaches for the quantification of {admission, discharge Dx} discrepancies do not specifically examine a subset of diseases or a specific system, but rather focus on identifying the overall effect on outcomes or patient profiles where {admission, discharge Dx} discrepancies are more common [6-8]. To our knowledge, no studies have been conducted with focus on respiratory conditions. Considering that respiratory Dx’s often share common symptoms and clinical manifestations, they may pose a challenge to clinical decision makers who correctly diagnose the patient at the point of admission. For this reason, we established an informed hypothesis that it is likely for respiratory Dx patients to be assigned with the incorrect admission Dx code. While respiratory diseases often share similar symptoms, there are specific attributes and small details that may help clinical decision makers distinguish between them, during differential diagnosis. Table 1 compares shared and disease-specific symptoms for COPD, asthma and pneumonia, as outlined by Mayo Clinic [9]. The thorough and non-sporadic use of diagnostic protocols and differential diagnosis tools are very important for improved diagnosis accuracy [10, 11]. In addition, the seasonal aspect of diagnosis and misdiagnosis in hospital, is reasonable to be examined, considering how seasonality plays a role in the prevalence of many respiratory diseases.

Table 1.

Comparison of symptoms between asthma, COPD and pneumonia

	Shortness of breath	Difficulty breathing	Wheezing	Chest pain	Cough	Disease-Specific Symptoms
Asthma	[x]	[x]	[x] *when exhaling	[x] *or tightness	[x] *in attacks	Trouble sleeping due to shortness of breath, coughing or wheezing
COPD	[x] *esp. during physical activities		[x]	[x] *chest tightness	[x] *chronic, may produce mucus	Frequent respiratory infections, lack of energy, weight loss (later stages), swelling (ankles/feet/ legs), need to clear throat in the morning due to excess mucus in lungs
Pneumonia	[x]	[x]		[x]*during breathing or cough	[x] *often phlegm producing	Confusion/changes in mental awareness (age> 65), fatigue, fever, sweating, chills, low temp. (age>65 and with weak immune systems), nausea, vomiting or diarrhea

Source: Mayo Clinic Diseases and Conditions online database (https://www.mayoclinic.org/diseases-conditions/index)

The first objective of this research is to measure the degree of discrepancies between the admission and the discharge Dx’s for respiratory patients, and the degree that this initial misdiagnosis leads to increased hospital charges and prolonged hospital stay. Since a number of respiratory conditions share similar diagnostic criteria and are often confused with each other, we specifically investigated how patients are initially labelled with incorrect respiratory Dx’s, other than the correct one that would later be assigned on discharge. We also want to examine how the seasonal variation of these discrepancies vary between winter and summer months. The second objective is the development of an exemplar desktop utility that integrates research results to show to the user the probability for these discrepancies, when a presumed admission Dx is provided as an input, and provide differential diagnosis information, accordingly.

This research aims to raise awareness on the importance of an evidence based and robust diagnosis triage process upon admission, and the uninterrupted and thorough examination of up-to-date diagnosis protocols during the clinical encounter, to increase the probability for the assigned Dx to match the discharge diagnosis. Such an accomplishment is anticipated to have significant implications on patient safety and the cost of care. There are numerous examples and reports regarding the importance of correctly diagnosing respiratory conditions, without delay. In a case study, a patient who visited the ED with cough, sweating, chest pain, was misdiagnosed with pneumonia and then was discharged. Days later he returned to the ED, coughing up blood and died from a delayed tuberculosis diagnosis. [12].

For the purpose of this study we transformed ICD-9 codes to Clinical Classification Software (CCS) codes developed by the Agency for Healthcare Research and Quality (AHRQ) [13], and studied all CCS categories of respiratory conditions, as classified by CCS. These conditions are asthma, COPD, bronchitis, pneumonia, tuberculosis, other lower respiratory infections. We decided to add the CCS category viral infections due to its often respiratory-related clinical manifestations. We also introduced two measures: Length of Stay (LOS) and Claims Payment Amount to study how the misdiagnosis of respiratory diseases is associated with increased hospital charges and prolonged hospital stay.

Terminologies

International Classification of Diseases: The International Classification of Diseases (ICD) is the nomenclature system for diseases standardized by World Health Organization (WHO) for reporting diseases and health conditions. Different diseases, injuries, disorders and other medical conditions are classified under ICD. The ICD version has different revisions, most recent being ICD 10. Both the admission and primary discharge Dx attributes in our dataset were coded using the 9th ICD revision [14]. There are more than 14,000 unique codes in ICD-9.

Clinical Classification of Software Codes (CCS): CCS classifies each ICD-9 code into broader disease categories. Since there are more than 14,000 different ICD-9 codes, we used the single level CCS to group the ICD Codes into a smaller number of exclusive categories of diseases. The CCS to ICD-9 mapping is available from the Healthcare Cost and Utilization Project (HCUP), which is a federal-state industry partnership sponsored by AHRQ [13]. The cardinality of the relationship between the ICD and CCS is N-1. For example: ‘481’ is the ICD 9 Code for ‘Pneumonia due to Streptococcus pneumoniae’ while ‘483’ is ICD 9 Code for ‘Pneumonia due to other specified organism’. Both would be grouped under the same CCS code (CCS=122). To transform our diagnosis attributes from the ICD to the CCS coding system, we used ICD2CCS, which is a standard crosswalk shared across medical organizations for this purpose.

Methodology

Data Sources

We have used the SynPUF dataset that is publicly available from the Center for Medicare and Medicaid Services (CMS). This is a synthetic medical claims dataset that simulates real hospital admissions data but with blinded patient information. The data in SynPUF files include the same patterns and trends that can be found in non-synthetic datasets. These SynPUF datasets are made available to facilitate research effort while providing protection to sensitive personal health information. CMS makes available 20 different subsets of SynPUF data. We used the Inpatient Claims files of SynPUF, since our focus are the inpatient admissions, so as to study the nature of discrepancies between final and admission Dx for respiratory patients. Every sample consisted of 3 years of data for patients admitted between 2008 and 2010. These are the most recent SynPUF data made available by CMS. Each sample consisted of approximately 65,000 records and as we merged 10 datasets, the final number of records was approximately 650,000 records. The attributes that we used for this research are described below [15].

Admitting ICD-9 Diagnosis Code: This is the initial Dx code on the institutional claim indicating the beneficiary’s initial diagnosis at the time of hospital admission, before any further patient investigation took place. In this paper we refer to this attribute as AD (Admission Diagnosis).

ICD-9 Diagnosis Code 1: The main discharge Dx code identifying the beneficiary’s principal diagnosis. While each beneficiary can have up to 10 different Dx codes in the dataset, positions 2 through 10 represent other comorbidities besides the beneficiary’s primary Dx code. The principal diagnosis typically represents the health problem that caused the need for hospitalization. In this paper, we refer to this attribute as PDD (Principal Discharge Diagnosis).

Claims Admission Date: The inpatient Admission Date in YYYYMMDD format. This field refers to the date the beneficiary was admitted to the hospital or skilled nursing facility.

Beneficiary Discharge Date: The date when the patient was discharged from hospital after the end of the treatment. This variable, combined with the aforementioned Claims Admission Date, was used to estimate the hospital length of stay per beneficiary.

Claim Payment Amount: The amount of payment made from the Medicare trust fund for the services covered by the claim record. Generally, the amount is calculated by the FI or carrier and represents what was paid to the institutional provider, physician, or supplier.

Approach

Based on the admission date, the dataset was split into two subsets, one including admissions that occurred during the three summer months (June 1st through August 31st) and the second only including admissions that occurred during the three winter months (December 1st through February 28th/29th). Each analysis was conducted twice, on the ‘Winter’ and the ‘Summer’ dataset respectively.

The two target datasets were firstly used for the calculation of conditional probabilities for AD → PDD pairs of respiratory conditions, for the summer and for the winter season. For instance, the formula: P (PDD=Pneumonia | AD=Asthma) is the probability that the PDD is pneumonia when an AD of asthma is initially assigned. For example, for ‘asthma’, an array of conditional probabilities as shown below are calculated:

{P (PDD=Asthma | AD=Asthma)} (a)

{P (PDD=Pneumonia | AD=Asthma),

P (PDD=COPD | AD=Asthma),

P (PDD=Bronchitis | AD=Asthma),

P (PDD=Tuberculosis | AD=Asthma),

P (PDD=Other low resp. infection | AD=Asthma),

P (PDD=Viral Infection | AD=Asthma),

P (PDD=Other conditions | AD=Asthma)} (b)

{P (PDD=non-Asthma | AD=Asthma)} (c)

The result of (a) is the probability that AD of asthma is the correct one (matching the PDD). On the other hand, the results in array (b) are seven probabilities for the AD of asthma to finally be found, on discharge, to be one of the seven other conditions under study. A concatenated probability was finally calculated as shown in (c) which is the probability for the AD of asthma to be anything other than asthma (overall mismatch). Similar arrays of probabilities were calculated for all seven respiratory conditions under study. This collection of conditional probabilities provides information about the probability for an admission Dx to be correct.

We also calculated the reverse conditional probabilities for the PDDà AD pairs under study. Examples for asthma are shown in (d), (e) and (f). These are the retrospective probabilities of ADs, given a known PDD. These probabilities inform us “What physicians initially thought when they were trying to diagnose a -later known-discharge diagnosis”.

{P (AD=Asthma | PDD=Asthma)} (d)

{P (AD=Pneumonia | PDD=Asthma),

P (AD=COPD | PDD=Asthma),

P (AD=Bronchitis | PDD=Asthma),

P (AD=Tuberculosis | PDD=Asthma),

P (AD=Other low resp. infection | PDD=Asthma),

P (AD=Viral Infection | PDD=Asthma),

P (AD=Other conditions | PDD=Asthma)} (e)

{P (PDD=non-Asthma | PDD=Asthma)} (f)

The next phase of the study was to estimate the mean and standard deviation of the length of stay and the hospital charges, for each of the aforementioned AD→ PDD pairs, so as to compare AD→ PDD discrepancies vs no discrepancies for the length of stay and the hospital charges. We furthermore conducted independent sample t-test analysis to examine the statistical significance of any differences.

The results of our research were integrated into an SQL database. The database was used as the backend of a desktop utility that was developed using .net desktop app development framework. The application allows the user to select an AD and see the probability for this input to lead to a matching or a different respiratory PDDs. The user can opt to choose any of these candidate PDDs in order to see diagnostic criteria and recommendations that can be used to reassess the patient before confirming the admission diagnosis. We used the online Mayo clinic database of diseases and conditions [9] in order to organize the diagnosis criteria and develop the recommendations component. Figure 1 is an outline of our overall methodological framework.

Figure 1.

Methodological framework of the study

Results

P (PDD| AD)

“Probability for the correct discharge diagnosis”

We found that approximately 25% of the patients that were thought to have asthma on admission are finally diagnosed with asthma on discharge. This percent was slightly higher for the winter admissions (Table 1). On the contrary, almost half of patients who were thought to have asthma on admission, were actually diagnosed with COPD upon discharge (44.708% in the winter and 42.809% in the summer). Furthermore, two thirds of patients that were believed to have COPD on admission would be actually be diagnosed with COPD on discharge. This percent is similar in both seasons. It was also found that approximately 12% of patients thought to have COPD on admission, would be eventually be with asthma upon discharge. Similar to COPD, almost two thirds of the patients who were believed to have pneumonia on admission were finally diagnosed with pneumonia on discharge. A significant proportion of patients that were believed to have pneumonia, on admission, were actually found to have another, non-respiratory condition, as discharge diagnosis. The rates for all discrepancies are presented on Table 2.

Table 2.

Seasonal comparison of correct and incorrect PDDs for the ADs of pneumonia, asthma, and COPD (AD → PDD)

P (AD | PDD)

“Whatphysicians initially thought when they were trying to diagnose a Dx”.

This is the reverse conditional probability, which quantifies the retrospective probability of admission Dx’s, given a known primary discharge Dx.

According to results, only a small proportion of patients with a PDD of asthma were initially identified as asthmatic on admission (11.419% of winter and 11.061% of summer admissions). On the contrary, about one third of patients with a PDD of asthma, were thought to have COPD on admission (33.703% of the winter and 30.941% of the summer admissions). Additionally, approximately 27% of patients with a PDD of asthma, were thought to have a low respiratory infection, on admission. Half of the beneficiaries, in both seasons, with a PDD of pneumonia had been initially identified with pneumonia on the admission phase. On the other hand, about 18% of beneficiaries with a PDD of asthma, were thought to have the CCS code of ‘Other Low Respiratory Infection’ on admission.

Finally, for COPD, approximately 45% of patients with a discharge Dx of COPD, were assigned with an admission Dx of COPD, on admission. A significant proportion of beneficiaries discharged with COPD, were thought to have another low respiratory infection on admission (28.522% for summer vs 29.211% for winter admissions). Finally, as it can be seen on Table 3, it is worth mentioning that of the three examined respiratory conditions, the PDD of pneumonia had the highest % of ADs of other non-respiratory conditions during the admission. Table 3 shows the rates for all discrepancies.

Table 3.

Seasonal comparison of correct and incorrect ADs when the PDD is pneumonia, asthma, and COPD

The bubble graphs below visualize the AD and PDD discrepancies for the three respiratory conditions asthma, pneumonia and COPD. Each bubble radius is analogous to the respective proportions. Clearly, from the graphs, COPD is confused with asthma very frequently. A PDD of asthma is rarely recognized correctly on admission. Also, on admission when decision makers believe that the Dx is asthma, they are proven incorrect, since they later on assign COPD as PDD on a significant portion of those beneficiaries. The bubble graphs also show that discrepancies are smaller when the AD or PDD of interest is COPD and pneumonia.

Comparison of mean LOS between matching and non-matching {AD, PDD} pairs

Differences were found to the mean LOS between matching and non-matching {AD, PDD} pairs, for the majority of the respiratory conditions under study. In those cases where the PDD of asthma, was correctly found upon admission, the hospital LOS was almost 0.7 days lower from cases where the PDD of asthma was mistaken as non-asthma on admission, with the difference being bigger for the winter admissions. As it is shown on Table 4, in the case on pneumonia, the LOS difference is even bigger, reaching almost one less day of hospital stay when pneumonia is correctly identified upon admission. Independent sample t-test was also conducted to test the statistical significance of the aforementioned differences. As shown on Table 4, the LOS differences between matching and non-matching {AD, PDD} pairs, were found to be statistically significant at the level of 0.05, for the PDDs of asthma, COPD, Other Low Respiratory Infections, pneumonia, and viral infections.

Table 4.

Comparison between matching and non-matching {admission, discharge Dx} pairs in terms of LOS and charges during winter

Comparison of hospital charges between matching and non-matching {admission, discharge Dx} pairs

Statistically significant differences were also observed to the mean hospital charges between matching and non-matching {AD, PDD} pairs, for most of the respiratory conditions under study. Specifically, for asthma, COPD, and pneumonia the total hospital charges are significantly lower when the Dx was correctly set at the time of the admission. The biggest difference was found to be in the case of asthma, and pneumonia, where the charges are on average $1,495 and $2,488 higher, respectively, when those PDDs matched the initial AD. As shown on Table 4, the differences in hospital charges between matching and non-matching {AD, PDD} pairs, were found to be statistically significant at the level of 0.05, for the PPDs of asthma, COPD, pneumonia, and viral infections.

Admission Dx Recommender Utility

A utility was developed that integrates the conditional probability P (PDD | AD) calculations, to serve as a recommender tool for clinical decision makers. The utility is intended to be used during the process of selecting an AD for the patient. By using the utility, physicians may select a respiratory AD that they believe their patient should be assigned to, on admission, and then they are informed about the probability for the selected Dx to be the correct one (matching the PDD), as well as the probabilities for mismatch with different respiratory PDDs. The utility was developed with the .net desktop app development framework.

Once the results are shown on screen, the physician is furthermore given the option to choose any of these candidate PDDs in order to see diagnostic criteria and recommendations that can be used to reassess the patient before confirming the AD. These diagnosis criteria and differential Dx information is specific to the {AD, PDD} under investigation. For instance, if the physician selects asthma as AD, he/she will learn about the increased probability for the PDD to be COPD, therefore reconsidering to review the patient information again. The system will provide {asthma, COPD}-specific differential diagnosis knowledge. Figure 3 shows the system workflow with the user interaction elements.

Figure 3.

The functionality and interaction of the clinical decision maker with the prototype utility

We used the online Mayo clinic database was used [9] in order to organize the diagnosis criteria and develop the recommendations component. Figure 4 shows one scenario of the application functionality. The authors acknowledge that this utility is not intended to be used as a verification system, in its current format. This is because there is a multitude of other parameters of care, such as the patient demographics, other clinical information and laboratory examination results which are not taken into account during the recommendation. Since the utility is probability based, the inclusion of additional parameters would increase its value and contextual usefulness.

Figure 4.

The user selects an AD to see the probability for this Dx to be the correct one, matching PDD or not (left image). Then the user can select any other candidate PDD to read pair-specific differential diagnosis information (right image).

Discussion and Conclusion

This study examined differences between ADs and PDDs for respiratory conditions and compared these differences in terms of the length of stay and hospital charges. It was found that asthma is very frequently mistaken for COPD or other low respiratory infections. A total of 60% of patients who received a PDD of asthma had initially been assigned with an AD of either COPD or the CCS code ‘Other Low Respiratory Infections’. COPD is characterized by decreased airflow over time, and inflammation of the tissues that line the airway. Asthma is a separate respiratory disease, often mistaken for COPD. The two conditions have similar symptoms, such as shortness of breath, cough, and wheezing, as well as quite similar comorbidities. It should be noted, though, that the two conditions have different causes and triggers. Especially, for the latter, asthma is made worse by exposure to allergens, cold air and exercise, while COPD aggravations are caused by respiratory tract infections and exposure to environmental pollutants. Other differences and diagnostic criteria do exist. We also found that COPD is frequently misclassified as ‘Other Low Respiratory Infection’ on admission. These two conditions are also of similar nature, but COPD is a chronic while the latter is an acute condition. Health systems are expected to be able to differentiate between these two candidate primary conditions, on admission.

We compared, for each respiratory condition, the mean LOS between correct (AD matching PDD) vs incorrect diagnosis (AD-PDD mismatch). For all conditions under study, the LOS was lower when the PDD was set correctly, early, on admission. This difference was found to be up to 1.2 days, in the case of pneumonia.

After examining the statistical significance of the differences, using t-test, we found the LOS difference to be significant in the case of asthma, COPD, pneumonia, and the CCS code ‘Other Low Respiratory Infection’. Findings were similar when we compared the correct and incorrect Dx’s in terms of the hospital charges. The mean hospital charges were found to be almost 40% higher when asthma and pneumonia were misdiagnosed as different conditions on discharge. After examining the statistical significance of the differences, using t-test, we found the total charges difference to be significant in the case of asthma, COPD, and pneumonia.

We also found that the observed patterns of discrepancies, were similar in winter and for summer admissions. An explanation may be that the diagnosis challenges are not affected positively or negatively by any prevalence differences of some respiratory conditions between the two seasons. Differential diagnosis challenges still remain similar.

This was a secondary analysis with the use of medical claims data; our approach stresses out the importance of clinical and administrative data reuse in health analytics projects to understand patterns of care and to find inconsistencies or areas for improvements in terms of decision making. The authors believe that by utilizing the admission and discharge diagnosis information with this inherent uncertainty in mind, future clinical decision support system designs will provide more contextually relevant insights: For instance the admitting diagnosis information in diagnostic predictive models should be used cautiously as model input, in order to avoid ‘Historical Decision’ biases [16].

While the results of this research add to existing knowledge that there are significant discrepancies between admission and discharge diagnosis in hospitals, they also shed light on the quality characteristics of these discrepancies for respiratory conditions. Clinical decision makers who assign admission Dx’s and hospital administrators, should be aware that the correct early Dx identification, matching the final diagnosis, evidently translates to savings and reduced hospital stay. Admission diagnosis verification systems such as the exemplar application presented in this paper should be included to the functionality of clinical decision support systems and should not be limited to respiratory conditions. Those systems can integrate discrepancy-specific differential diagnosis information [17], in a fashion similar to our approach. For instance, physicians who select an admission Dx code that often leads to a different discharge Dx, would be presented with differential diagnosis resources and protocols that would pinpoint to aspects, patient history considerations and clinical manifestations that may be further examined. It is also important for future work to consider the reasons why decision makers opt for specific admission diagnoses over other ones, in a patterned way (such as in the asthma-COPD mismatch), and whether these patterns are a result of a systematic hospital-level approach to initial diagnosing, or simply consistent errors due to similarities in symptomatology. The latter stresses the importance and value for robust differential diagnosis early on, during the hospital stay.

Hospitals may also consider conducting cost-benefit analyses in order to consider investing on more thorough initial patient assessment systems and flows. Investment considerations may include the following four components: (i) the recruitment of specialty physicians for teleconsultation sessions during the initial patient assessment [18, 19], (ii) the availability of state of the art diagnostic equipment to improve the quality of diagnosis (iii) the integration of differential diagnosis protocols and verification systems to existing Electronic Medical Record Systems, (iv) development of advanced data analytics methods [20], which will model the patient demographics, patient history and clinical manifestations in order to provide assistive diagnosis output. Medical education and training should also be factored in, during these efforts.

Figure 2.

Top Left: A small proportion of patients with a PDD of Asthma were identified as asthmatic on admission. Bottom left: Most patients identified as asthmatic on admission were finally diagnosed with COPD. Top middle: Most patients with a PDD of pneumonia were identified with pneumonia at the admission phase. Bottom middle: Most patients that were thought to have pneumonia upon admission were later actually diagnosed with pneumonia. Top right: Almost half of patients with a PDD of COPD were correctly identified as COPD patients on admission. Bottom right: Most patients identified with COPD upon admission were later found to be COPD patients, in most cases.