Little data exist on HCV phylodynamics and transmission networks among people who inject drugs (PWID), especially from low- and middle-income countries (LMICs). HCV epidemics in a city can be considered a series of sub-epidemics caused by phylogenetically distinct viral lineages. Mapping these lineages to generate transmission clusters and overlaying epidemiologic data can be used to identify factors associated with clustering. PWID were recruited via respondent driven sampling in 2016-17. Participants completed a survey and blood draw. HCV 5’UTR-core sequencing was performed on 486 HCV RNA positive samples from 4 cities (Amritsar [n=126], Delhi [n=128], Kanpur [n=138], Imphal [n=94]). Sequences were aligned using Multiple Sequence Comparison by Log-Expectation. The most appropriate nucleotide substitution model was determined using jModelTest and phylogenetic inference was carried out using Maximum Likelihood methods in RaXML with 500 bootstrap replications. Clusters were identified using ClusterPicker with posterior support and genetic distance thresholds of 70% and 4.5%, respectively. Given the large number of covariates of interest, a machine learning model utilizing the Boruta wrapper of the random forest algorithm was constructed to identify features predictive of clustering, as well as differences between clusters. Median age was 33 years, 99% were male and HIV prevalence was 75%. Mean p-distance for all sequences was 0.075. A total of 251 sequences fell into 19 transmission clusters (Fig). Mean cluster size was 7.4 (range: 2-49); 8 clusters were dyads. There were 6 large clusters comprised of >10 samples. 7 of the 19 clusters contained samples from multiple cities. Machine learning based analysis revealed that no history of HIV testing and living with friends were predictive of clustering (both p<0.05), and that state, residential zip code, injection zip code, time spent away from home, and buprenorphine injection could be predictive of membership in a given cluster (all p<0.05). Age, gender, and HIV status did not predict clustering. These are among the first data from a LMIC setting to demonstrate clustering across multiple cities. The median size of the clusters identified were also larger than self-reported injection networks in India. Treatment as prevention efforts for HCV have emphasized network-based approaches for PWID, and these data suggest that networks may need to be defined by space (zip code) as opposed to egocentric injection networks.