For some $c \in \mathbb{F}_p^*$ consider the polynomial $$ f(t) = t^3 - 3ct^2 - 3t + c $$ for $p \equiv 1$ (mod $3$) and $p \equiv 3$ (mod $4$). In this case $3$ is a quadratic non-residue modulo $p$ and hence the discriminant $\Delta = 2^23^3(c^2+1)^2$ of $f$ is also a quadratic non-residue modulo $p$.
Thus $f$ has exactly one root $r \in \mathbb{F}_p$. For $c=1$ the root $r = -1$. How explicitly express $r$ through $c$ in general case without usage of the Cardano formula (i.e., without cubic roots)?
It is indeed possible to express the root as a rational function of $\ c\ $. This follows from the fact that$^{\mathbf{\dagger}}$, under the stated conditions, the polynomial $\ t-r\ $ must be the gcd of the polynomials $\ f\left(t\right)\ $ and $\ t^{p-1}-1\ $. In the Euclidean algorithm for computing the gcd, the coefficients of the initial polynomials $\ t^{p-1}-1\ $ and $\ f\left(t\right)\ $ are rational functions of $\ c\ $ (with coefficients in $\mathbb F_p\ $), and it's easy enough to show that the remainder of a polynomial $\ \pi_1(t)\in \mathbb F_p[t]\ $ modulo another, $\ \pi_2(t)\ $, has coefficients that are rational functions of those of $\ \pi_1(t)\ $ and $\ \pi_2(t)\ $. From this it follows that the coefficients of the gcd must be rational functions of those of the initial polynomials, and hence of $\ c\ $.
In practice, the function can be found for any given $\ p\ $ by treating $\ c\ $ as an indeterminate, and calculating the gcd in $\ \mathbb F_p(c)[t]\ $, where $\ \mathbb F_p(c)\ $ is the field of rational functions in $\ c\ $ over $\ \mathbb F_p\ $. Since $\ c\ $, as an element of $\ \mathbb F_p^*\ $, is a root of the polynomial $\ t^{p-1}-1\ $, monomials, $\ c^q\ $, of degree $\ p-1\ $ or higher can nevertheless be replaced by $\ c^{q\,\mathrm{mod}\,(p-1)}\ $ as the calculation proceeds. The denominator of the rational function obtained by this procedure cannot have a linear factor other than $\ c\ $, since if it did, being $\ c-\rho\ $, say, with $\ \rho \ne 0\ $, then the degree of $\ \gcd\left( t^{p-1} - 1, t^3 - 3\rho t^2 - 3t^2 + \rho\right)\ $ would have to be zero.
Here are the expressions for $\ r\ $ for the cases $\ p=7,19\ $ and $\ 31\ $, respectively:
$$\frac{3}{5c^5+5c^3+c}\\ \ \\ \frac{\small 4c^{14}+3c^{12}+10c^{10}+12c^8+14c^6+4c^4+8c^2+7}{\small 17c^{17}+17c^{15}+18c^{13}+4c^{11}+2c^9+11c^7+17c^5+13c^3+10c}\\ \ \\ \frac{\tiny 15c^{26}+5c^{24}+5c^{22}+16c^{20}+21c^{18}+2c^{16 }+29c^{14}+29c^{12}+24c^{10}+20c^8+5c^6+14c^4+14c^2+11}{\tiny 29c^{29}+29c^{27}+17c^{25}+5c^{23}+18c^{21}+17c^{19}+2c^{17}+25c^{15}+12c^{13}+9c^{11}+12c^7+3c^5+12c^3+3c}$$
The pattern observable here in the leading terms of the numerators and denominators continues through the primes $\ 43\ $ and $\ 67\ $, the first term of the numerator being $\ (p-4)\,c^{p-5}\ $ in all cases except $\ p=7\ $, and the first two terms of the denominator being $\ (p-2)\,c^{p-2} + (p-2)\,c^{p-4}\ $ in all cases.
The softwear package I used to perform the gcd calculations crashed when I tried $\ p=79\ $, so I haven't looked at any primes beyond $\ 67\ $.
$\mathbf\dagger$ Or even from from the simple fact that it's a function of $\ c\ $, as Eric Wofsey points out in the comments below.
Update: As Eric Wofsey observes in the comment below, the root can also be expressed as a polynomial function of $\ c\ $. If we let $\ c_r = \frac{r^3-3\,r}{3\,r^2-1}\ \mathrm{mod}\ p\ $, the value of $\ c\ $ corresponding to the root $\ r\ $, one well-known expression for such a polynomial is: $$\sum_{r\in {\mathbb F}_p^*} r \prod_{u\in{\mathbb F}_p^*\setminus\{c_r\}} \frac{c-u}{r-u}\ .$$ For $\ p=7,19\ $ and $\ 31\ $ the polynomials are: $$ 5\,c^5 + 4\,c^3 + 4\,c \\ \ \\ 8\,c^{17}+6\,c^{13}+4\,c^{11}+2\,{c^9}+7\,c^7+16\,c^5+5\,c^3+8\,c\\ \ \\ {\tiny 3c^{29}+22c^{27}+5c^{25}+12c^{23}+12c^{21}+15c^{19}+27c^{17}+5c^{15}+16c^{11}+14c^9+3c^7+22c^5+17c^3+12c} $$