Thread by @rheum_cat

Brooke McGinley (mentor @mikelavalley8) with @bu_bmc_rheum @tuhina_neogi

Identifying subgroups of pts undergoing knee replacement in #MOST

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi 🔹In the US more than 640,000 knee replacements annually

🔹In the US estimated 10.2 billion in annual costs

🔹Knee osteoarthritis consists of different phenotypes often leading to a knee replacement outcome

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi Purpose:

To cluster patients who have undergone knee replacement into SUBGROUPS according to demographic and clinical characteristics to develop PHENOTYPES for use in clinical management of knee #OA

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi Data from OAI

Sample: Underwent knee replacement any time during the 9 years of follow up

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi From OAI, these variables were collected

red = collected at baseline

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi Used 2 unsupervised learning approaches:

1️⃣PCA
2️⃣k means clustering

Lots of variables to consider ➡reduce dimensions

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi 📈PCA: Combine potential predictors into a small number of composite variables (PCA dimensions)

📈K means clustering: Create K distinct and non-overlapping clusters

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi n=414, mean age 68, 60% female

PCA: 3 dimensions

Dim 1: female sex, BMI
Dim 2: WOMAC scores
Dim 3: analgesic use, OA severity

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi K means clustering:

Cluster 1: older women, low BMI, Heberden’s nodes
Cluster 2: men, higher weight, height, phys activity
Cluster 3: h/o knee injury or surgery, low use of opioids
Cluster 4: younger women, high BMI, pain, disability, depression scores.

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi 👉 Found clear clusters along dimensions of sex, age, history of knee injury or surgery, and race with differences in activity, pain, disability, and depression among clusters

#ACR22

@MikeLaValley8 @BU_BMC_Rheum @Tuhina_Neogi 👉Clinical implications: Offers the promise of more tailored clinical management of OA patients

👉Clusters require further validation before use in clinical management

#ACR22