grep invert search with context

Give this a try:

sed 'h;:b;$b;N;N;/PATTERN/{N;d};$b;P;D' inputfile

You can vary the number of N commands before the pattern to affect the number of lines to delete.

You could programmatically build a string containing the number of N commands:

C=2 # corresponds to grep -C
N=N
for ((i = 0; i < C - 1; i++)); do N=$N";N"; done
sed "h;:b;\$b;$N;/PATTERN/{N;d};\$b;P;D" inputfile

If the lines are all unique you could grep the lines you want to remove into a file, and then use that file to remove the lines from the original, e.g.

grep -C 2 "line I don't want" < A.txt > B.txt
grep -f B.txt A.txt

awk 'BEGIN{n=2}{a[++i]=$0}
/dont/{
  for(j=1;j<=i-(n+1);j++)print a[j];
  for(o=1;o<=n;o++)getline;
  delete a}
END{for(i in a)print a[i]} ' file