Using latent class analysis to identify subgroups of postoperative older adults

Abstract

Objective

Identify subgroups of postoperative older adults using a latent class analysis (LCA) of routinely collected patient data, including preoperative frailty and postoperative mobility.

Methods

Retrospective analysis of routinely collected data for adults 65 years and older (N=2,036) admitted to the hospital following surgery at our institution. We utilized LCA to identify patient subgroups based on routinely collected measures of frailty, mobility, activities of daily living, and general health status. We compared hospital outcomes between the identified groups, including extended length of stay (LOS) (>0.5 SD beyond mean LOS by surgical category), discharge disposition (i.e., home versus non-home discharge), and utilization (weekly visit frequency) of physical therapy (PT) and occupational therapy (OT).

Results

We identified 3 subgroups of patients that we labeled Low Frailty-High Mobility (LF-HM), High Frailty-Low Mobility (HF-LM), and Low Frailty-Low Mobility (LF-LM), representing 15.3%, 27.6%, and 57.1% of the cohort, respectively. Discharge to home was highest among the LF-HM group (99%), followed by LF-LM (96%), and HF-LM (77%). Extended LOS was most common among the HF-LM group (27%), followed by LF-LM (18%), and LF-HM (6%). PT and OT visit frequencies were highest in the HF-LM group followed by the LF-LM and LF-HM groups.

Conclusions

This study identified 3 subgroups of postoperative older adults using routinely collected patient data. These groups may help to explain variations in hospital outcomes that have not been adequately explained using measures of preoperative or postoperative measures alone. These results may be useful in developing tailored postoperative care for older adults.

Keywords

geriatric surgery precision medicine electronic health record data health services research

Introduction

More than half of the 20 million surgeries performed each year in the US are for adults who will require an inpatient hospital stay.¹ Among these, older adults are at increased risk for extended hospital lengths of stay (LOS) and discharge to post-acute care facilities,^2–7 which conflict with patient preferences and contribute to increased cost of care.^8–10 Previous studies have identified individual patient-level factors associated with postoperative outcomes. For example, studies have found that preoperative frailty is associated with increased risk for loss of independence following surgery, as well as longer LOS, among older adults.^7,11 Another study found that patients with lower levels of mobility one day after gastrointestinal surgery were more likely to experience an extended LOS.¹² Additional studies have observed associations between age, gender, body mass index, preoperative function and American Society of Anesthesiologists (ASA) rating with postoperative outcomes following surgery.^2,6,13,14 However, few studies have examined preoperative and postoperative patient data in combination, which is likely to account for variability in patient outcomes better than individual preoperative or postoperative measures alone.

Latent class analysis (LCA) is a model-based statistical approach used to identify hidden subgroups (i.e., “latent classes”) within a population based on patterns across multiple observed variables.¹⁵ As a type of finite mixture modeling, LCA identifies groups of individuals who share similar characteristics or response patterns, allowing researchers to characterize heterogeneity that may not be directly observable. Identifying subgroups in this manner is commonly used to inform the development of tailored or “precision” treatment approaches.^16–20 The purpose of this study was to identify subgroups of postoperative older adults using a latent class analysis of routinely collected patient-level data from a large academic health care system. We also examined differences in key postoperative outcomes among those in each of the identified latent subgroups.

Methods

We utilized a retrospective cohort design to identify latent subgroups of older adults admitted to the hospital following surgery and to examine differences in hospital outcomes among those in each latent subgroup. Data for this study was collected during routine care encounters and extracted from the electronic health record (EHR) for analysis.

Patient cohort

We included older adults (≥65) hospitalized at Johns Hopkins Hospital or Johns Hopkins Bayview Medical Center following inpatient surgery between September 2016 and March 2020. Eligible patients included those with data available for each of our class indicators (described below) and having any of the following surgical procedures: abdominal surgery, gynecologic surgery, pancreatic surgery, head & neck surgery, neurosurgery, orthopaedic surgery, plastic surgery, spine surgery, thoracic surgery, urologic surgery, or vascular surgery. Individuals undergoing outpatient surgery were not included in our cohort.

Latent class indicators

Based on previous evidence and standard data collection practices at our institution, we utilized patient age, general health status, frailty, mobility, and activities of daily living to identify latent subgroups of older adults after surgery.^{2,6,7,11–14} General health status was measured prior to surgery by the anesthesiology team using the American Society of Anesthesiologists Physical Status Classification System (ASA PS Classification), a method used to succinctly assess and communicate patients’ pre-anesthesia medical comorbidities.²¹ ASA PS Classification scores range from 1 (normal health patient) to 6 (brain-dead). Frailty was measured at preoperative appointments by nursing using the Edmonton Frail Scale (EFS), a valid and reliable measure of frailty with scores ranging from 0-17 and higher scores indicative of greater frailty.^22,23 Patient mobility function was assessed using the Activity Measure for Post-Acute Care (AM-PAC) Inpatient Basic Mobility “6-Clicks” Short Form.^24,25 Patient function in activities of daily living was measured using the AM-PAC Inpatient Daily Activity “6-Clicks” Short Form. Both AM-PAC short forms include 6 items scored by clinicians on a scale of 1-4 with lower scores indicating greater impairment. Raw scores were converted to t-scores and used in all analyses; these t-scores range from 16.6 to 57.7.²⁶ In our hospitals, AM-PAC short forms are scored daily by nursing and at all physical therapy (PT) visits (mobility short form) and occupational therapy (OT) visits (daily activity short form). Previous studies have shown high interrater reliability across and within disciplines for these measures.^27,28 We utilized the first observed AM-PAC mobility and daily activity scores recorded by either discipline following surgery (maximum 48 hours after surgery) for analysis.

Hospital outcomes

Hospital outcomes for this study included LOS, discharge disposition, and utilization of PT and OT services. LOS was categorized as an extended LOS, defined as a LOS (days) that was ≥0.5 standard deviations above the mean LOS for each surgical category, or a non-extended LOS, defined as any LOS less than the extended LOS definition. This approach was used to define extended LOS as LOS varies substantially depending on the type of surgery a patient receives. Defining extended LOS in this manner accounts for differences between surgical categories while still allowing our team to examine the number of patients experiencing an extended LOS across the full cohort. Discharge disposition was categorized as home discharge or non-home discharge (e.g., skilled nursing facility, inpatient rehabilitation facility). PT and OT utilization were examined separately based on weekly visit frequency, which was identified based on the number of PT or OT visits received, divided by patients’ LOS, and multiplied by 7.

Statistical analysis

Descriptive statistics were used to describe the sample. Latent class analysis was used to identify unobserved subgroups within the study population based on patterns across multiple class indicator variables. This model-based clustering approach estimates the probability that individuals belong to distinct latent classes defined by similar combinations of indicator values. Class indicators used in this study included a combination of continuous and categorical variables. Continuous class indicators included: age, AM-PAC mobility and AM-PAC daily activity scores, and EFS total score. Categorical class indicators included ASA ratings and individual EFS items (i.e., general health status, functional dependence, functional performance). ASA ratings were dichotomized to those patients who were healthy or with mild systemic disease (i.e., scores of 1-2) or patients with severe disease (i.e., scores of 3 and above). The general health status item on the EFS was dichotomized to those patients in good to excellent health or patients in fair to poor health. The functional dependence criterion of the EFS was dichotomized to those who required help in 0-1 activities or those who required help in 2 or more activities. The functional performance criterion of the EFS was dichotomized to those who completed the timed up and go test in 0-10 seconds or for those who completed it in 11 or more seconds. All LCA models were estimated using Mplus, Version 8.6.²⁹ Fit statistics, classification accuracy and relative class size were considered in selecting the best fitting latent class model. Guidelines suggest a minimum of 500 observations for LCA as used in this study.¹⁵

Model fit was measured using the Bayesian information criterion (BIC) and the sample size-adjusted BIC (aBIC).³⁰ For the BIC and aBIC, smaller values indicate better model fit. The Vuong-Lo-Mendell-Rubin Likelihood Ratio Test (VLMR-LRT) was used to compare the relative fit of a model with k classes to a model with one fewer class. Significant p values for the VLMR-LRT indicate that the latent class solution with fewer classes should be rejected in support of the solution with more classes.³⁰ Entropy, ranging from 0 to 1, was also examined with larger values indicating better classification accuracy. Finally, class sizes were examined for each solution as simulation studies indicate that small or uncommon classes can be difficult to identify reliably and it is important to avoid over extracting classes.³⁰

Once the unconditional latent class model was established (step 1), we used the auxiliary function in Mplus with Bolck, Croon and Hagenaars (BCH) weights to examine differences across classes in the prevalence of two dichotomous distal hospital outcomes (i.e., discharge disposition, extended length of stay) and in the means of two continuous distal hospital outcomes (i.e., PT and OT weekly frequencies). In this automatic 2-step BCH approach, weights were applied to individual participants based on their posterior probabilities of class membership to account for classification uncertainty and an overall test of equality as well as pairwise tests of differences in each outcome across classes using one degree of freedom were conducted.^29,31 P-values of less than 0.01 were considered statistically significant.

The surgical heterogeneity present in our sample has the potential to obscure meaningful patterns among more homogenous surgical populations within our cohort. To examine the influence of including patients with multiple types of surgeries in our analysis, we conducted two sensitivity analyses that limited our analysis to patients with spine surgery or abdominal surgery. These populations were selected as they represented homogenous patient groups and were among the largest subgroups.

Results

We identified 3,497 patients who completed the EFS prior to surgery between September 2016 and March 2020. Among these patients, we identified 2,036 with complete AM-PAC mobility and daily activity scores recorded within 48 hours of surgery. ASA ratings and other class indicators were present for all patients with EFS and AM-PAC scores. Our cohort was 45% female and the majority of patients were White (76.6%) or Black (16.1%). The average length of stay among included patients was 4.4 days (SD: 4.4). Length of stay varied by surgical category, from 2.8 days (SD: 3.3) in urologic surgery to 8.5 days (SD: 5.7) in pancreatic/hepatobiliary surgery. The most common surgical categories were abdominal surgery (21.2%) and urologic surgery (20.3%) (Table 1).

Table 1.

Cohort description.

Age, mean (SD)	73.0 (5.9)
Female, No. (%)	917 (45.0)
Race, No. (%)
White	1559 (76.6)
Black	328 (16.1)
Asian	70 (3.4)
Other	79 (3.9)
Ethnicity, No. (%)
Hispanic	46 (2.3)
Non-Hispanic	1973 (96.9)
Other/unknown	17 (0.8)
EFS Total Score, mean (SD)	3.43 (2.6)
AMPAC Mobility t-score, mean (SD)	42.4 (7.7)
AMPAC Daily activity t-score, mean (SD)	42.0 (7.7)
ASA Physical Status^a, No. (%)
1	2 (0.1)
2	570 (28.0)
3	1388 (68.2)
4	76 (3.7)
Health Status^b, No. (%)
0	1301 (63.9)
1	649 (31.9)
2	86 (4.2)
Functional Dependence^c, No. (%)
0	1714 (84.2)
1	217 (10.7)
2	105 (5.2)
Functional Performance^d, No. (%)
0	1157 (56.8)
1	669 (32.9)
2	210 (10.3)
Length of stay (days), mean (SD)	4.4 (4.4)
Surgical specialty, No. (%)
Abdominal	432 (21.2)
Gynecology	89 (4.4)
Hepatobiliary/pancreatic	239 (11.7)
Head and neck	59 (2.9)
Neurosurgery	194 (9.5)
Orthopaedics	100 (4.9)
Plastics	65 (3.2)
Spine	287 (14.1)
Thoracic	99 (4.9)
Urology	414 (20.3)
Vascular	58 (2.8)
Total, No.	2036

^aASA Ratings for those in this study included 1 (normal healthy patient), 2 (patient with mild systemic disease without significant functional limitation or end-organ involvement), 3 (patient with one or more severe systemic diseases causing substantive functional limitation) and 4 (patient with severe systemic disease that poses a constant threat to life).

^bBased on number of hospitalizations over past year and self-reported general health (higher scores = worse health, instrument score range 0-2).

^cBased on the number of daily activities that an individual requires assistance with (higher scores= more dependence, instrument score range 0-2).

^dBased on impact of baseline health on ability to complete physical activity (higher scores = more impaired function, instrument score range 0-2).

EFS: Edmonton Frail Scale; AM-PAC: Activity Measure for Post-Acute Care; ASA: American Society of Anesthesiologists.

For our LCA, comparison of the fit statistics, classification quality, and class size for solutions specifying one to eight classes suggested several reasonable models (see Table 2). Although the eight-class model had the lowest BIC and aBIC values, the scree plot indicated that minimal benefit of extending the model beyond five- or six-classes (visualized as the line flattening between the models with five or six classes indicating minimal decrease in BIC and aBIC for models with more than 6 classes). The VLMR indicated that a four-class model fit the data better than a three-class model and a five-class model fit better than a four-class model but adding a sixth class did not improve model fit. However, the smallest class in the four-class model included only 2-3% of the sample (33-57 people) suggesting that these patients were outliers on each of the measures, and not a particularly clinically meaningful class. Based on all of these considerations, we selected the three-class model as the best fitting model and further analyses focused on that solution.

Table 2.

Fit indices for latent class models.

# Classes	LL-value	# of free parameters	BIC	aBIC	VLMR-LRT	VLMR-LRT p value	Entropy	Smallest class n (%)
1	-30253	12	60597	60559	---	---	---	---
2	-29179	21	58518	58452	2147.6	0.00	0.80	630 (30.9)
3	-28718	30	57664	57568	923.2	0.00	0.83	311 (15.3)
4	-28319	39	56936	56812	796.3	0.00	0.88	57 (2.8)
5	-28108	48	56583	56430	421.9	0.00	0.84	48 (2.3)
6	-27935	57	56305	56124	346.00	0.40	0.85	34 (1.7)
7	-27808	66	56119	55910	254.34	0.12	0.85	34 (1.7)
8	-27706	75	55983	55745	204.98	0.15	0.85	33 (1.6)

Note. LL-value = Log-likelihood value. BIC = Bayesian information criterion. aBIC = sample size adjusted BIC. LMR = Lo-Mendell-Rubin likelihood ratio test. LMR-LRT and entropy are not applicable for the 1-class model.

Based on the characteristics of the members of each of the identified classes, we labeled the 3 identified classes as the Low Frailty-High Mobility (LF-HM) class (15.3%), the High Frailty-Low Mobility (HF-LM) class (27.6%), and the Low Frailty-Low Mobility (LF-LM) class (57.1%) (Table 3). Demographics, which were not used to generate classes, varied by class with the highest proportion of females in the HF-LM class (54.4%) and lowest proportion of females in the LF-HM class (35.5%). Race also varied by class with a higher proportion of Black patients in the HF-LM class (21.1%) compared to the LF-LM class (13.5%).

Table 3.

Description of identified classes and outcomes of class membership.

		Low frailty-high mobility (15.3%)	High frailty-low mobility (27.6%)	Low frailty-low mobility (57.1%)
Demographics	Sex
	Female	105 (35.5%)	302 (54.4%)	510 (43.0%)
	Male	191 (64.5%)	253 (45.6%)	675 (57.0%)
	Race
	White	231 (78.0%)	376 (67.7%)	952 (80.3%)
	Black	51 (17.2%)	117 (21.1%)	160 (13.5%)
	Asian	6 (2.0%)	29 (5.2%)	35 (3.0%)
	Other	8 (2.7%)	33 (5.9%)	38 (3.2%)
	Ethnicity,
	Hispanic	6 (2.0%)	16 (2.9%)	24 (2.0%)
	Non-Hispanic	289 (97.6%)	533 (96.0%)	1151 (97.1%)
	Other/unknown	1 (0.3%)	6 (1.1%)	10 (0.8%)
Class Indicators	Age	71.59	75.03	72.46
	EFS Total Score, mean	2.57	6.46	2.20
	AMPAC Mobility t-score, mean	54.24	37.68	41.59
	AMPAC Daily activity t-score, mean	54.78	37.53	40.68
	ASA Rating 1-2, (%)	0.36	0.13	0.33
	ASA Rating 3-4, (%)	0.64	0.87	0.67
	Health status -0, (%)	0.75	0.27	0.79
	Health status -1-2, (%)	0.25	0.73	0.21
	Functional dependence -0, (%)	0.96 ^α	0.49	0.98
	Functional dependence -1-2, (%)	0.05 ^α	0.51	0.02
	Functional performance -0, (%)	0.70	0.15	0.74
	Functional performance -1-2, (%)	0.30	0.85	0.26
Hospital Outcomes	Discharge to home, % (SE)	99 (1)†	77 (2)†	96 (1)†
	Extended length of stay*, % (SE)	6 (2)†	27 (2)†	18 (1)†
	Physical therapy visits (per week), mean (SE)	0.3 (0.07)†	2.1 (0.10)†	1.2 (0.06)†
	Occupational therapy visits (per week), mean (SE)	0.1 (0.04)†	1.1 (0.07)†	0.5 (0.04)†

N=2,036.

Note. For categorical hospital outcomes, values by class represent proportions. For continuous outcomes values represent means.

†Significantly different compared to other classes in row (p<0.001).

^αDo not sum to 100% due to rounding.

EFS: Edmonton Frail Scale; AM-PAC: Activity Measure for Post-Acute Care; ASA: American Society of Anesthesiologists.

*Represents a length of stay ≥0.5 SD above mean length of stay (days), by surgical category.

There were notable differences in class indicators between the three classes (Table 3). Age was similar between the LF-HM and LF-LM classes (71.6 and 72.5 years, respectively), but was higher in the HF-LM class (75.0 years). Frailty scores were similar in the LF-HM class (2.6) and LF-LM class (2.2) but were notably higher among those in the HF-LM class. Mean AM-PAC mobility scores were highest in the LF-HM class (54.2), followed by the LF-LM class (41.6), and HF-LM class (37.7). AM-PAC daily activity scores followed a similar pattern with mean scores of 54.8, 40.7, and 37.5 in the LF-HM, LF-LM, and HF-LM classes, respectively. ASA also differed by class with the HF-LM class having the highest probability (0.87) of receiving an ASA rating of 3 or higher and the LF-HM and LF-LM having similar but lower probabilities of receiving an ASA rating of 3 or higher (0.64 and 0.67, respectively). Individual EFS items also differed by class. Specifically, individuals in the HF-LM had a much higher probability of having fair to poor health status, being functionally dependent, and requiring more than 10 seconds to complete the up and go test as compared to the LF-HM and LF-LM classes.

Class membership varied by surgical category (Figure 1). For example, the LF-HM class was most represented among those undergoing head & neck surgery (32%) and was least represented among those undergoing hepatobiliary/pancreatic surgery (2%). The HF-LM class was most represented among those undergoing vascular surgery (50%) and least represented by those undergoing urologic surgery (12%). Finally, the LF-LM class was most represented by those undergoing hepatobiliary/pancreatic surgery (69%) and least among those undergoing plastic surgery (38%).

Figure 1.

Class Membership by Surgical Category HB: hepatobiliary.

All hospital outcomes of interest were significantly different by class (p<0.001) (Table 3). The proportion of those discharged home was 99% in the LF-HM class, 96% in the LF-LM class, and 77% in the HF-LM class. We observed that the percentage of patients experiencing an extended LOS was highest in the HF-LM class (27%), followed by the LF-LM class (18%), and the LF-HM class (6%). PT and OT visit frequencies were highest in HF-LM class (PT: 2.1 visits/week; OT: 1.1 visits/week), followed by the LF-LM class (PT: 1.2 visits/week; OT: 0.5 visit/week), and the LF-HM class (PT: 0.3 visits/week; OT: 0.1 visits/week).

We conducted two sensitivity analyses where we repeated each step of our LCA among¹ patients who had abdominal surgery (n=437) and² patients who had spine surgery (n=287). For the abdominal surgery group, we identified a 3-class model with results that were very similar to the full cohort with EFS scores within 1 point and AM-PAC scores within 2 points of those reported for the full cohort. For the spine surgery group, we also selected a 3-class model. While this model was not statistically superior to the 2-class model based on the VLMR p-value, we viewed this as a product of low sample size versus poor model fit, especially because other fit statistics, including smallest class size (12.1%), supported this model. This model was similar to the 3-class model identified using the full cohort with some notable differences. The spine HF-LM class included higher frailty scores than the equivalent class in the full cohort model. We also observed lower mobility scores across all 3 classes compared to the full cohort. Taken together, these sensitivity analyses suggest that the 3-class model identified using the full cohort is valid across different surgical categories but that the levels of frailty, mobility, and other class indicators may vary slightly by surgical category. Complete results of the sensitivity analyses are available as Supplemental Material.

Discussion

The results of this study indicate that LCA may be a useful approach for identifying latent subgroups of postoperative older adults, a large patient population vulnerable to poor outcomes. We specifically identified three latent subgroups of postoperative older adults, which we have labeled as Low Frailty-High Mobility (LF-HM) (15.3%), High Frailty-Low Mobility (HF-LM) (27.6%), and Low-Frailty-Low Mobility (LF-LM) (57.1%). These subgroups were present within each of the included surgical categories, although the proportion of patients belonging to each subgroup varied by type of surgery. Furthermore, patients in each subgroup differed significantly from one another in discharge disposition, LOS, and utilization of PT and OT services.

The classes identified in this study align with clinical reasoning and highlight the importance of physical function in the International Classification of Functioning, Disability and Health (ICF) domains of mobility and self-care (activities of daily living).³² It was not surprising to us that our analysis identified a subgroup of patients (HF-LM) with poorer general health, higher frailty scores, and lower physical function and that these patients were more commonly discharged to post-acute care and experienced higher rates of extended LOS. Likewise, it stands to reason that our analysis identified another class of patients (LF-HM) that were generally healthy, had low frailty scores, had high mobility and daily activity function scores, were nearly always discharged home and rarely experienced an extended LOS. However, less expected was the identification of the LF-LM class, which is similar to the LF-HM group for all class identifiers, with the exception of AM-PAC scores which are both lower in the LF-LM class compared to the LF-HM class. Despite only this difference between groups, the LF-LM class experienced extended LOS rates 3 times higher and utilized 4-5 times as many PT and OT visits per week compared to the LF-HM class. These findings align with previous studies that have highlighted the influence of patient mobility and daily activity function on hospital outcomes and the importance of measuring these items as part of routine care.^12,33–35

The findings of this study may inform strategies for developing tailored postoperative care for postoperative older adults by providing insights into how preoperative and postoperative factors may influence patient outcomes in combination. For example, previous studies suggest immediate postoperative mobility levels are closely tied with LOS and discharge disposition.^12,33 In this study, we identified two groups of patients with relatively low mobility score (i.e., HF-LM, LF-LM). However, the group with higher preoperative frailty experienced extended LOS and non-home discharges significantly more often than those with lower frailty – suggesting that both of these measures should be considered when determining what resources (e.g., PT, OT) patients may require during their postoperative recovery. While the results of this study do immediately provide a deployable classification tool or new treatment strategy, they do suggest that tailored or “precision” postoperative care strategies for older adults should account for both preoperative frailty and postoperative mobility as these factors provide a more comprehensive view of the patient than either measure in isolation.

The classes identified in this study, while corresponding to one institution, are likely generalizable to other institutions performing surgery among older adults seeking to identify clinical subgroups. Importantly, this study was conducted exclusively using routinely collected data extracted from the EHR. While this increases the overall feasibility of the study performed, it does place additional emphasis on the identification and selection of clinical measures, as well as systematic collection and documentation of these measures. Previous publications can provide guidance on the process of selecting outcome measures for healthcare systems seeking to implement or refine data collection strategies.^36,37

Strengths of this study include a large patient sample, well beyond the sample size of ≥500 observations recommended for LCA studies.¹⁵ This study also utilized data elements that are routinely collected at many hospitals, increasing the generalizability of our results. There are also limitations to this study that should be considered when interpreting results. Our study used a retrospective design and was limited to data available in our EHR. It is likely that there are factors not measured routinely in our healthcare system that would have altered the subgroups identified in this study had they been included in our analysis. For example, social determinants of health (e.g., caregiver availability, home accessibility, family support structures) and mental health (i.e., depression) have been shown to influence LOS and discharge disposition among hospitalized patients and may have altered the composition of the classes we identified.^38–42 Our analysis was also performed specifically among older adults after surgery and it is possible that this affects the generalizability of our results to other populations, such as postoperative adults under 65 years of age. It is also possible that missing data that affected eligibility for our analysis (e.g., AM-PAC) was not missing at random and could have systematically biased our results in ways unknown to the authors, affecting generalizability.

Conclusion

Using LCA to explore subgroups of postoperative adults, we identified 3 subgroups of patients based on routinely collected data extracted from the EHR. We observed different proportions of patients with extended LOS and discharged home across the identified subgroups. We also observed different rates of PT and OT utilization among these groups. Our results suggest that combining preoperative and postoperative measures may better account for heterogeneity of outcomes among postoperative older adults than preoperative or postoperative measures along. Using these routinely collected measures may help to inform care decisions following surgery for older adults. However, further research is needed to validate the findings of our analysis at other institutions, as well as to examine the influence of social factors on the classes identified in this study and the hospital outcomes observed among these groups.

Supplemental material

Supplemental material - Using latent class analysis to identify subgroups of postoperative older adults

Supplemental material for Using latent class analysis to identify subgroups of postoperative older adults by Kevin McLaughlin, Amie Bettencourt, Daniel Young, Erik Hoyer, Michael Friedman, Elizabeth Colantuoni, Lee A. Goeddel and Pedro Gozalo by Sage Open Medicine.

Footnotes

ORCID iD

Kevin H. McLaughlin

Ethical considerations

This study was acknowledged as exempt from review by the Johns Hopkins Medicine Institutional Review Board (IRB00337243).

Consent to participate

A waiver of written informed consent was formally granted by the Johns Hopkins Medicine Institutional Review Board.

Author contributions

Conceptualization: KM, DY, EH, MF, LG. Methodology: EC, AB. Data curation: KM, EH, LG. Formal Analysis: AB, KM. Writing (original draft): KM, AB, PG. Writing (reviewing and editing): KM, AB, DY, EH, MF, EC, LG, PG.

Funding

The authors disclosed receipt of the following financial support for the research, authorship, and/or publication of this article: Funding for this study was provided by the Learning Health Systems Rehabilitation Research Network (LeaRRn) (NIH 5P2CHD101895-03)

Declaration of conflicting interests

The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Preprint

An earlier version of this manuscript is available as a preprint at

Supplemental material

Supplemental material for this article is available online.

References

Inpatient vs. Outpatient Surgeries

[Internet]. [cited 2021 Oct 5]. Available from: https://www.ahrq.gov/data/infographics/outpatient-surgery.html

Collins

Daley

Henderson

, et al. Risk Factors for Prolonged Length of Stay After Major Elective Surgery. Ann Surg 1999; 230(2): 251–259, PubMed PMID: 10450740; PubMed Central PMCID: PMC1420868. https://doi.org/10.1097/00000658-199908000-00016

Clark

Ryan

. Concurrent Prediction of Hospital Mortality and Length of Stay from Risk Factors on Admission. Health Services Research 2002; 37(3): 631–645. https://doi.org/10.1111/1475-6773.00041

Cleveland Clinic Orthopaedic Arthroplasty Group . The Main Predictors of Length of Stay After Total Knee Arthroplasty: Patient-Related or Procedure-Related Risk Factors. JBJS 2019; 101(12): 1093–1101. https://doi.org/10.2106/JBJS.18.00758

Karhade

Ogink

Thio

QCBS

, et al. Discharge Disposition After Anterior Cervical Discectomy and Fusion. World Neurosurgery 2019; 132: e14–e20. https://doi.org/10.1016/j.wneu.2019.09.026

Al-Mazrou

Chiuzan

Kiran

. Factors influencing discharge disposition after colectomy. Surg Endosc 2018; 32(7): 3032–3040. https://doi.org/10.1007/s00464-017-6013-z

Goeddel

Murphy

Owodunni

, et al. Domains of Frailty Predict Loss of Independence in Older Adults After Noncardiac Surgery. Annals of Surgery 2023; 278(2): e226–e233. https://doi.org/10.1097/SLA.0000000000005720

Henry

Halpin

Hunt

, et al. Patient disposition and long-term outcomes after valve surgery in octogenarians. Ann Thorac Surg 2012; 94(3): 744–750. https://doi.org/10.1016/j.athoracsur.2012.04.073

Lala

Chang

Liu

, et al. Risk for non-home discharge following surgery for ischemic mitral valve disease. J Thorac Cardiovasc Surg 2020; S0022-5223(20): 30524–30529, PubMed PMID: 32307181; PubMed Central PMCID: PMC7874266. https://doi.org/10.1016/j.jtcvs.2020.02.084

10.

Okoh

Haik

Singh

, et al. Discharge disposition of older patients undergoing trans-catheter aortic valve replacement and its impact on survival. Catheter Cardiovasc Interv 2019; 94(3): 448–455. https://doi.org/10.1002/ccd.28069

11.

Mrdutt

Papaconstantinou

Robinson

, et al. Preoperative Frailty and Surgical Outcomes Across Diverse Surgical Subspecialties in a Large Health Care System. Journal of the American College of Surgeons 2019; 228(4): 482–490. https://doi.org/10.1016/j.jamcollsurg.2018.12.036

12.

Carroll

Hampton

Carroll

, et al. Mobility scores as a predictor of length of stay in general surgery: a prospective cohort study. ANZ Journal of Surgery 2018; 88(9): 860–864. https://doi.org/10.1111/ans.14555

13.

Chen

Scholten

Chomsky-Higgins

, et al. Risk Factors Associated With Perioperative Complications and Prolonged Length of Stay After Laparoscopic Adrenalectomy. JAMA Surgery 2018; 153(11): 1036–1041. https://doi.org/10.1001/jamasurg.2018.2648

14.

Shah

Memon

Kay

, et al. Preoperative Patient Factors Affecting Length of Stay following Total Knee Arthroplasty: A Systematic Review and Meta-Analysis. The Journal of Arthroplasty 2019; 34(9): 2124–2165.e1. https://doi.org/10.1016/j.arth.2019.04.048

15.

Kongsted

Nielsen

. Latent Class Analysis in health research. Journal of Physiotherapy 2017; 63(1): 55–58. https://doi.org/10.1016/j.jphys.2016.05.018

16.

Grant

McCloskey

Hatfield

, et al. Use of Latent Class Analysis and k-Means Clustering to Identify Complex Patient Profiles. JAMA Network Open 2020; 3(12): e2029068. https://doi.org/10.1001/jamanetworkopen.2020.29068

17.

Mori

Krumholz

Allore

. Using Latent Class Analysis to Identify Hidden Clinical Phenotypes. JAMA 2020; 324(7): 700–701. https://doi.org/10.1001/jama.2020.2278

18.

Dandis

Westeneng

Inthout

, et al. Latent Class Analysis to Predict Outcomes of Early High-Intensity Physical Therapy After Total Knee Arthroplasty, Based on Longitudinal Trajectories of Walking Speed. J Orthop Sports Phys Ther 2021; 51(7): 362–371. https://doi.org/10.2519/jospt.2021.10299

19.

Ulbricht

Chrysanthopoulou

Levin

, et al. The use of latent class analysis for identifying subtypes of depression: A systematic review. Psychiatry Research 2018; 266: 228–246. https://doi.org/10.1016/j.psychres.2018.03.003

20.

Neumann

Wirtz

Ernstmann

, et al. Identifying and predicting subgroups of information needs among cancer patients: an initial study using latent class analysis. Support Care Cancer 2011; 19(8): 1197–1209. https://doi.org/10.1007/s00520-010-0939-1

21.

American Society of Anesthesiologists [Internet]. [cited 2024 Apr 23] . Statement on ASA Physical Status Classification System. Available from. https://www.asahq.org/standards-and-practice-parameters/statement-on-asa-physical-status-classification-system

22.

Hao

, et al. Assessment of predictive validity and feasibility of Edmonton Frail Scale in identifying postoperative complications among elderly patients: a prospective observational study. Sci Rep 2020; 10(1): 14682. https://doi.org/10.1038/s41598-020-71140-5

23.

Rolfson

Majumdar

Tsuyuki

, et al. Validity and reliability of the Edmonton Frail Scale. Age Ageing 2006; 35(5): 526–529, PubMed PMID: 16757522; PubMed Central PMCID: PMC5955195. https://doi.org/10.1093/ageing/afl041

24.

Tevald

Clancy

Butler

, et al. AM-PAC “6-Clicks” for the Prediction of Short Term Clinical Outcomes in Individuals Hospitalized with COVID-19: A Retrospective Cohort Study. Archives of Physical Medicine and Rehabilitation 2021; 0(0). https://doi.org/10.1016/j.apmr.2021.08.006

25.

Hull

Thut

. A Simple Tool Using AM-PAC “6-Clicks” to Measure Value Added in Acute Care Physical Therapy: The Therapy Value Quotient. Journal of Acute Care Physical Therapy 2018; 9(4): 155–162. https://doi.org/10.1097/JAT.0000000000000082

26.

Haley

Coster

Andres

, et al. Score comparability of short forms and computerized adaptive testing: simulation study with the activity measure for post-acute care 1. Archives of Physical Medicine and Rehabilitation 2004; 85(4): 661–666. https://doi.org/10.1016/j.apmr.2003.08.097

27.

Jette

Stilphen

Ranganathan

, et al. Interrater Reliability of AM-PAC “6-Clicks” Basic Mobility and Daily Activity Short Forms. Physical therapy 2015; 95(5): 758–766. https://doi.org/10.2522/ptj.20140174

28.

Hoyer

Young

Klein

, et al. Toward a Common Language for Measuring Patient Mobility in the Hospital: Reliability and Construct Validity of interprofessional Mobility Measures. Phys Ther 2017; 2: 133–142. https://doi.org/10.1093/ptj/pzx110

29.

Muthen

. Mplus User’s Guide. Muthen & Muthen. Eighth.

30.

Nylund

Asparouhov

Muthén

. Deciding on the Number of Classes in Latent Class Analysis and Growth Mixture Modeling: A Monte Carlo Simulation Study. Structural Equation Modeling: A Multidisciplinary Journal 2007; 14(4): 535–569. https://doi.org/10.1080/10705510701575396

31.

Asparouhov

Muthén

. Auxiliary variables in mixture modeling: Using the BCH method in Mplus to estimate a distal outcome model and an arbitrary secondary model. [Internet][cited 2024 Apr 26]. Available from. https://www.statmodel.com/examples/webnotes/webnote21.pdf (2016).

32.

Boonen

Maksymowych

. Measurement: function and mobility (focussing on the ICF framework). Best Pract Res Clin Rheumatol 2010; 24(5): 605–624. https://doi.org/10.1016/j.berh.2010.05.008

33.

Menendez

Schumacher

Ring

, et al.

Does “6-Clicks” Day 1 Postoperative Mobility Score Predict Discharge Disposition After Total Hip and Knee Arthroplasties?

The Journal of Arthroplasty 2016; 31(9): 1916–1920. https://doi.org/10.1016/j.arth.2016.02.017

34.

Blackwood

Fernandez

. AM-PAC 6-Clicks Scores Predict Hospital Discharge Destination In Older Adults With Cardiovascular Disease (poster). Innov Aging 2018; 2(Suppl 1): 489, PubMed PMID: null; PubMed Central PMCID: PMC6228152. https://doi.org/10.1093/geroni/igy023.1824

35.

McLaughlin

Friedman

Hoyer

, et al. The Johns Hopkins Activity and Mobility Promotion Program: A Framework to Increase Activity and Mobility Among Hospitalized Patients. Journal of Nursing Care Quality 2023; 38(2): 164–170. https://doi.org/10.1097/NCQ.0000000000000678

36.

Bull

Teede

Watson

, et al. Selecting and Implementing Patient-Reported Outcome and Experience Measures to Assess Health System Performance. JAMA Health Forum 2022; 3(4): e220326. https://doi.org/10.1001/jamahealthforum.2022.0326

37.

Franklin

Chenok

Lavalee

, et al. Framework To Guide The Collection And Use Of Patient-Reported Outcome Measures In The Learning Healthcare System. EGEMS (Wash DC) 2017; 5(1): 17, PubMed PMID: 29881737; PubMed Central PMCID: PMC5983040. https://doi.org/10.5334/egems.227

38.

Delanois

Tarazi

Wilkie

, et al.

Social determinants of health in total knee arthroplasty: are social factors associated with increased 30-day post-discharge cost of care and length of stay?

The Bone & Joint Journal 2021; 103-B(6 Supple A): 113–118, 103-B. https://doi.org/10.1302/0301-620X.103B6.BJJ-2020-2430.R1

39.

Segar

Keshvani

Rao

, et al. Race, Social Determinants of Health, and Length of Stay Among Hospitalized Patients With Heart Failure: An Analysis From the Get With The Guidelines-Heart Failure Registry. Circulation: Heart Failure 2022; 15(11): e009401. https://doi.org/10.1161/CIRCHEARTFAILURE.121.009401

40.

Nguyen

Page

Aggarwal

, et al. Social Determinants of Discharge Destination for Patients After Stroke With Low Admission FIM Instrument Scores. Archives of Physical Medicine and Rehabilitation 2007; 88(6): 740–744. https://doi.org/10.1016/j.apmr.2007.03.011

41.

Beeler

Cheetham

Held

, et al. Depression is independently associated with increased length of stay and readmissions in multimorbid inpatients. European Journal of Internal Medicine 2020; 73: 59–66. https://doi.org/10.1016/j.ejim.2019.11.012

42.

Gold

Slover

. Diagnosis of depression and other patient factors impacts length of stay after total knee arthroplasty. Arthroplasty Today 2020; 6(1): 77–80. https://doi.org/10.1016/j.artd.2019.11.010

Supplementary Material

Please find the following supplemental material available below.

For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.

For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.

0.05 MB

0.00 MB