Figure 16 of https://www.onsemi.com/pub/Collateral/2N3903-D.PDF indicates if you interpolate between 10 and 30mA to get 20mA , you can choose a saturation voltage of either 0.20 or 0.30V for Vce to get approximately 230 uA or 400uA of base current respectively.
Computed as a ratio of collector current and accounting for Vbe drop of 0.65 to 0.7 one gets an effective current gain of 20m/230u = 87 @ 0.3Vce to 20m/400uA = 50 @ 0.2Vce
The table specs use a defacto ratio of 10:1 Ic:Ib for Vce(sat) results for the lowest practical worst case (max) saturation voltage, but the chart shows a lower nominal Vce
So hFE=50 a maximum effective value @ Vce=0.2 V Rb=(5V-0.7{Vbe})/400uA= 10.8k Ohms
more
This proves ( as suggested in comments) that anything between 1k and 10k works with slight reduction in collector current from a rise in Vce. You can achieve a nominal linear hFE of 150 ONLY if Vce>=1V @ 10mA and perhaps 160 if Vce is higher. Yet hFE has a very high tolerance, so we use conservative values to get better results as a "switch".
Linear Hfe current gain is spec'd here for 100 to 300 @ 10mA @ Vce=1.0V in the tables for the 2N3904. These apply to generic supplier specs too.