Experimental data on compressive strength sigma(max) versus strain rate (epsilon) over dot(eng) for metallic glasses undergoing uniaxial compression show varying strain rate sensitivity. For some metallic glasses, sigma(max) decreases with increasing epsilon(eng), while for others, sigma(max) increases with increasing (epsilon) over dot(eng), and for certain alloys sigma(max) versus (epsilon) over dot(eng) is nonmonotonic. To understand their strain rate sensitivity, we conduct molecular dynamics simulations of metallic glasses undergoing uniaxial compression at finite strain rates and coupled to heat baths with a range of temperatures T-0 and damping parameters b. In the T-0 -> 0 and b -> 0 limits, we find that the compressive strength sigma(max) versus temperature T obeys a "chevron-shaped" scaling relation. In the low-strain-rate regime, sigma(max) decreases linearly with increasing T, whereas sigma(max) grows as a power law with decreasing T in the high-strain-rate regime. For T-0 > 0, sigma(max) (T) deviates from the scaling curve at low strain rates, but sigma(max) (T) rejoins the scaling curve as the strain rate increases. Enhanced dissipation reduces compression-induced heating, which causes sigma(max) (T) to deviate from the b -> 0 scaling behavior for intermediate strain rates, but sigma(max) (T) converges to the high-strain-rate power-law scaling behavior at sufficiently high strain rates. Determining sigma(max) (T) as a function of b and T-0 provides a general framework for explaining the strain rate sensitivity of metallic glasses under compression.