I am using a rank-ordered logit model to predict the winner of a multi-player game (think Fortnite). Among other individual-level factors (e.g., strength of each player, experience or number of games each player has played, etc.), I also have some historical data on past multi-player games, some of which might involve the players in the game I'm trying to predict. For example, today's game that I'm trying to predict has players A, B, C, D, & E of which from past history, I know that player A & B played in 3 of the same multi-player games with A scoring higher than B 2 out of 3 times.

My question is, what approach(es) might I consider to incorporate this matrix of past head to head results into my predictions for today's game? In many instances, the players in today's game haven't played in the same game with each other before, but in some instances they may have. Also, as noted, in addition to these head to head results, I also have individual-level data that I would like to incorporate in my prediction.

Thanks for your help!