You can implement three-input Toffoli gate with three two-input Toffoli gates and one ancilla qubit as shown below.

Assume that qubits $q_0$, $q_1$ and $q_2$ are three inputs and qubit $q_4$ is a target. Qubit $q_3$ is ancilla qubit. The first gate implements function $q_0~ \mathrm{AND}~ q_1$ and saves results of this operation to $q_3$. The second gate implements function $q_2~ \mathrm{AND}~ q_3 = q_2~\mathrm{AND}~(q_0~ \mathrm{AND}~ q_1) = q_0~\mathrm{AND}~q_1~ \mathrm{AND}~ q_2$. Eventually, qubits $q_0$, $q_1$ and $q_2$ control qubit $q_4$. The last gate is used for uncomputation ancilla qubit $q_3$ back to state $|0\rangle$ (this is necessary because ancilla qubits entangled with other qubits can cause interference and increase error rate, note that inverse gate for Toffoli is again Toffoli).
Here is a code in QASM (sorry, I am not experienced in Qiskit)
OPENQASM 2.0;
include "qelib1.inc";
qreg q[5];
creg c[5];
ccx q[0],q[1],q[3];
ccx q[2],q[3],q[4];
ccx q[0],q[1],q[3];