Development and Evaluation of a List of High-Risk Inpatient Procedures in Patients 65 Years and Older

Abstract

Background:

Inpatient procedures are common and important health events for older Americans. To facilitate surgical outcomes research, we sought to create and evaluate lists of International Classification of Disease, Tenth Revision (ICD-10) codes for high-risk inpatient procedures, defined as having at least a 1% inpatient mortality.

Methods:

This retrospective national cohort study analyzes Medicare claims from 2018 for patients 65 years and older undergoing inpatient procedures. Surgical Diagnosis Related Group (DRG) codes in the inpatient claims were used to identify procedures. We identified the primary ICD-10 procedure code for each patient and then compiled all codes with at least a 1% inpatient mortality yielding three separate lists: one list that was blind to elective vs. urgent/emergent status, and one each for urgent/emergent and elective procedures. Clinical review by three surgeons was used to remove procedures unlikely to be the proximate cause of mortality. For evaluation, we examined the mortality of each code among fee-for-service Medicare beneficiaries in 2017, 2019, and2020 to determine how many of these satisfied the 1% mortality criterion.

Results:

This study included 2,241,419 patients from 2018 undergoing inpatient procedures. The final result included 231 (blind to elective vs urgent/emergent status), 167 (urgent/emergent status), and 119 (elective status) ICD-10 procedure codes for the three lists. Our evaluation from 2017, 2019, and2020 demonstrated that in our master list, which was blind to elective vs. urgent/emergent status, 97.8% of procedures had an inpatient mortality of at least 1%. In our high-risk procedures lists for urgent/emergent and elective procedures, 100% and 94.1% of codes met this requirement.

Conclusions:

We developed and evaluated lists of ICD-10 codes representing high-risk procedures in patients 65 years and older. These lists will be powerful tools for researchers studying surgical outcomes.

Introduction

Inpatient procedures – conventional surgeries in addition to percutaneous procedures performed in catheterization labs or operating rooms – are important health events for older Americans. They are common in patients over 65 years¹ and carry risks for adverse events that have the potential to change patient function and independence. Studying inpatient procedures in contemporary databases frequently requires lists of International Classification of Disease, Tenth Revision (ICD-10) codes. A previous set of lists was developed by Schwarze et al. for patients older than 65 years using ICD-9 codes based on a data from a single state and then tested in a 20% sample of hospitals using the National Inpatient Sample². In this article we describe the development of updated lists of ICD-10 codes representing high-risk procedures through the use of national Medicare fee-for-service (FFS) claims data from 2018, and then evaluating using 2017, 2019, and2020 Medicare FFS claims data. Our three lists consisted of a master list blind to elective vs urgent/emergent status (“procedure status”) and one each for elective and urgent/emergent procedures.

Methods

Determining high-risk procedures

This study was approved by the institutional review board; individual patient consent was waived. To develop the high-risk lists, we examined all Medicare beneficiaries 65 years and older with FFS coverage in 2018 and undergoing an inpatient procedure, determined by Diagnosis Related Group (DRG) codes. We excluded organ transplant procedures. We identified the primary ICD-10 code and calculated the crude mortality by dividing the number of patients who died as inpatients by the total number of patients undergoing that procedure. As in previous work by Schwarze et al. 1% was used as the threshold for inclusion as a high-risk procedure². Three board-certified surgeons (SS, GS, MS) independently evaluated the lists to remove procedures that were unlikely to be the cause of death but that may occur in patients at high-risk for death. For example, tracheostomy creation is a procedure that is itself unlikely to be the cause of death but usually occurs in patients at high risk for death. Each surgeon independently assessed each procedure code. Disagreements were settled through discussion to reach a consensus. Last, contralateral procedures were added in cases where only one side appeared on the list and where risk was clinically unlikely to be related to laterality. For example, a left leg orthopedic procedure would be added if only the right leg procedure appeared on the list. We used this process to generate a master list blind to procedure status and one each for urgent/emergent and elective procedures.

Evaluation of the high-risk procedures list

To evaluate each of the three lists we calculated the mortality percentage for each code as described above in Medicare beneficiaries 65 years and older with FFS in 2017, 2019, and2020 using the corresponding procedure status.

Results

Development of High-Risk Lists

We identified 2,241,419 Medicare beneficiaries undergoing inpatient procedures in 2018. We identified procedures with a crude inpatient mortality of at least 1% to create a master list without regard to procedure status (i.e. urgent/emergent vs elective) of 494 distinct ICD-10 procedures (Figure 1). Restricting the list to urgent/emergent and elective procedures resulted in 354 and 205 procedures, respectively. After removal of procedures unlikely to be the cause of death as described above, 10, 6, and 10 contralateral procedures were added to the master, urgent/emergent, and elective lists. The final result was three lists – all patients irrespective of procedure status, urgent/emergent procedures, and elective procedures – with 231, 167, and 119 ICD-10 procedure codes each (Supplementary Tables 1-3). We have provided our lists of excluded procedures that met the 1% mortality threshold in Supplementary Table 4.

Figure 1.

Flow diagram demonstrating the stepwise generation of three lists of high-risk procedures: one each for urgent/emergent and elective procedures, and one for procedures irrespective of procedure status.

Evaluation of Lists

Of the 231 inpatient procedures in our master list, 226 (97.8%) had a mortality greater than or equal to 1% (Supplementary Table 1). Only 5 (2.2%) procedures fell below this threshold (Table 1). Of these 5 procedures, 1 was a procedure we had added manually to the original list as a contralateral procedure.

Table 1.

Procedures with at least 1% inpatient mortality in the development list but with less than 1% inpatient mortality in evaluation data

List	ICD-10	Mortality (%)	Procedure Description
Master List	02RF38Z	0.99	REPLACEMENT OF AORTIC VALVE WITH ZOOPLASTIC, PERC APPROACH
	0RG6071	0.98	FUSION THOR JT W AUTOL SUB, POST APPR P COL, OPEN
	0BTF4ZZ	0.89	RESECTION OF RIGHT LOWER LUNG LOBE, PERC ENDO APPROACH
	0DTG4ZZ	0.81	RESECTION OF LEFT LARGE INTESTINE, PERC ENDO APPROACH
Elective	0DTP0ZZ	0.93	RESECTION OF RECTUM, OPEN APPROACH
	0RG6071	0.80	FUSION THOR JT W AUTOL SUB, POST APPR P COL, OPEN
	0BTF4ZZ	0.76	RESECTION OF RIGHT LOWER LUNG LOBE, PERC ENDO APPROACH
	0RG1071	0.75	FUSION CERV JT W AUTOL SUB, POST APPR P COL, OPEN
	041L0JL	0.72	BYPASS L FEM ART TO POPLIT ART WITH SYNTH SUB, OPEN APPROACH

For the high-risk urgent/emergent and elective lists, 0 of 167 and 7 (5.9%) of 119 procedures had a mortality less than 1% (Table 1, Supplementary Tables 2, 3). In the latter group, two of the procedures had been manually added as contralateral procedures.

Discussion

Schwarze et al. derived lists of high-risk surgeries based on a Pennsylvania state database using International Classification of Disease, Ninth Revision (ICD-9) codes². We updated these lists for contemporary analysis by using ICD-10 codes in national Medicare data and expanded our procedures to include percutaneous procedures as well. Our decision to expand reflects the increasing frequency of these procedures^3,4 as well as the fact that a percutaneous approach or a catheterization lab setting is not a proxy for low-risk.

Our results suggest that the main list of high-risk inpatient procedures performed well, with 97.8% of these procedures having the requisite 1% or higher inpatient mortality in the 2017, 2019, and2020 FFS population. When excluding contralateral procedures that had not met this criterion in the derivation dataset, 217 (98.1%) of the procedures had at least a 1% inpatient mortality.

These lists will be tools for researchers seeking to study high-risk procedures as a whole or in groups (e.g. by anatomic location, specialty, or approach). The creation and evaluation of the high-risk procedures list arose from our group’s work on surgical/procedural outcomes in patients living with dementia. We believe it will be a useful tool for others seeking to understand outcomes on the basis of procedural risk.

Our approach to the development of high-risk lists has several weaknesses and strengths. It uses a crude mortality threshold rather that one adjusted for individual patient factors and for clustering by provider and facility. In other words, it does not disentangle the contributions to risk from the procedure, provider, facility, and patient. As a consequence it is not appropriate as an individual patient risk calculator but best serves the purpose outlined above. On a related note, distinguishing procedures related to mortality rather than those merely occurring in high-risk patients is an inherently subjective process. We used a rigorous iterative process with multiple surgeons, all with advanced clinical training and working at diverse institutions to evaluate procedures to produce the final lists. Nevertheless we recognize that a different set of individuals may disagree on some of the specific choices that we made. Also, there may some variation in whether particular procedures are performed as inpatient or outpatient services based on provider and institutional practices. We speculate that this would apply to a minority of cases and is unlikely to affect the overall value of our high-risk lists. Next, this study is based on analysis of an administrative dataset. Although ICD-10 codes are substantially more granular than ICD-9 codes, these still cannot fully capture the nuance of clinical practice. As a corollary point, these billing codes are not used perfectly accurately, which likely contributes to clinically implausible findings, such as discrepancies between equivalent procedures with different laterality. For example, 0Y6C0Z2, a type of right-sided above-knee amputation, had a mortality of 4.2% in the derivation data while the left-sided analogue was <1%. There are several strengths to our study. Our study is based on fee-for-service Medicare data, which is a rich national dataset with a large sample size. This has the effect of reducing regional variation impacting mortality. Our source data also makes our lists ideal for research among patients 65 years and older. Second, our list is based on contemporary data and ICD-10 rather than ICD-9 codes.

With regard to our approach to its evaluation, there are also several limitations. First, the evaluation itself was restricted to three years. Given that 94.1-100% of procedures in our testing dataset fit our desired mortality criterion, however, we do not believe that testing in additional years would yield results different enough to call the usefulness of the list into question. Second, the high-risk procedures list was tested only in Medicare fee-for-service samples and did not include Medicare part C enrollees. Extrapolation to our databases must be done cautiously but it is clinically unlikely that procedures deemed high-risk in Medicare would have meaningfully different results in alternative datasets. Third, we used a 1% inpatient mortality as our threshold for high-risk rather than alternatives, such as 15- or 30-day postoperative mortality. While we evaluated the performance of our codes across three full years of national data, there are two limitations: first, one of our years was 2020, the year of the COVID-19 pandemic, and second, we did not evaluate to see if cases that met the 1% mortality criterion in the evaluation data fell below it in the derivation data. As a consequence, it is likely that certain procedures fell below the 1% inpatient mortality threshold in our derivation sample but met or exceeded it in other years. While we do not believe either of these substantively reduces the usefulness of our lists, either of these factors could have impacted our results and so we have intentionally avoided labeling them as fully validated. As such, future efforts should focus on the evaluation of procedures excluded from our evaluation year that may have met the 1% threshold in other years.

Conclusion

We developed lists of ICD-10 procedure codes representing high-risk inpatient procedures in patients 65 years and older based on Medicare FFS data and using clinical consensus to remove procedures unlikely to have been the proximate cause of mortality. Our evaluation of the high-risk procedures lists confirms that they effectively select for inpatient procedures with at least a 1% mortality. Our lists can be used for studying high-risk inpatient procedures in Medicare FFS beneficiaries. Extrapolation to other datasets is clinically reasonable but should be done with caution.

Supplementary Material

Supinfo

Supplementary Material Title.

High-risk procedure lists with mortality in the derivation and evaluation data sets

Key Points.

• We created lists of International Classification of Disease, Tenth Revision (ICD-10) procedure codes representing high-risk inpatient procedures with at least a 1% inpatient mortality, including a master list of 231 codes blind to elective vs. urgent/emergent status.

• We evaluated our lists in 2017, 2019, and 2020 Medicare data and found that 226 (97.8%) procedures from this master list had the minimal 1% mortality.

Why does this paper matter?

This paper provides three lists of ICD-10 codes – one blind to elective vs urgent/emergent status and one each for elective and urgent/emergent procedures – representing high-risk inpatient procedures in patients 65 years and older. These lists will be invaluable research tools for those investigating surgical outcomes among older adults.