Here is one way of doing this problem that is a little different than the way you were going or the hint of @peter a g.
Let $D(\alpha)$ be the polynomial discriminant of the minimal polynomial for $\alpha$ and let $d(K)$ be the field discriminant for $K$.
The quantity $[\mathcal{O}_K : \mathbb{Z}[\alpha]]$ is sometimes called the index.
It turns out that there is a formula relating the discriminant of the minimal polynomial, the discriminant of the field, and the index:
$$ D(\alpha)=[\mathcal{O}_K : \mathbb{Z}[\alpha]]^2\cdot d(K). $$
Assuming that $a\in\mathbb{Q}$ we have that $x^n-a$ is the minimal polynomial for $\alpha$. In particular, since $p|a$ and $p^2\nmid a$ it follows that $x^n-a$ is $p$-Eisenstein. Now since the minimal polynomial of $\alpha$ is $p$-Eisenstein we have that $p$ is totally ramified in $\mathbb{Q}(\alpha)$, and more importantly for us that $p^{n-1}||d(K)$ if $p\nmid n$ and $p^n|d(K)$ if $p|n$ (this is theorem 3.6 here).
The polynomial discriminant of $x^n-a$ is known to be $D(\alpha)=\pm n^n a^{n-1}$.
So, when $p\nmid n$ we have $p^{n-1}||d(K)$ and that $p^{n-1}||D(\alpha)=\pm n^n a^{n-1}$ since $p\nmid n$ and $p||a$. But now by the `index-discriminant' formula
$$D(\alpha)=[\mathcal{O}_K : \mathbb{Z}[\alpha]]^2\cdot d(K) $$
and since $p^{n-1}||D(\alpha)$ and $p^{n-1}||d(K)$ it follows that $p\nmid [\mathcal{O}_K : \mathbb{Z}[\alpha]]$.
When $p|n$ things are a little less strait forward. One way to proceed is to use the following theorem which is equivalent to part of Theorem 6.1.4 in Cohen's book A Course in Computational Algebraic Number Theory:
Theorem (Dedekind) Let $\alpha$ be an algebraic integer with minimal
polynomial $m$ and set $K=\mathbb{Q}(\alpha)$. Let $p$ be a prime, and write
$$ m(x)=\prod_{i=1}^r m_i(x)^{e_i} \pmod{p} $$
where $m_i\in\mathbb{Z}[x]$ are
monic, irreducible lifts of the irreducible factors of $m$ modulo $p$.
Set $$ g(x)=\prod_{1\leq i\leq r} m_i(x),\hspace{3mm}
h(x)=\prod_{1\leq i\leq r} m_i(x)^{e_i-1}, \hspace{3mm} \text{and }\
f(x)=\frac{g(x)h(x)-m(x)}{p}. $$
Then $p|[\mathcal{O}_K : \mathbb{Z}[\alpha]]$ if and only if
$\gcd(\bar{f},\bar{g},\bar{h})\neq 1$, where over-lines denote
reduction modulo p.
In our situation we have
$$m(x)=x^n-a\equiv x^n\pmod{p}.$$
Therefore $g(x)=x$ and $h(x)= x^{n-1}$ and also
$$ f(x)=\frac{x^n-(x^n-a)}{p}.$$
So by Dedekind's Theorem we have $p|[\mathcal{O}_K : \mathbb{Z}[\alpha]]$ if and only if $\gcd(\bar{f},\bar{g},\bar{h})\neq 1$ which can only happen only if $x|\bar{f}$. However $x|\bar{f}$ only when $f(0)\equiv 0 \pmod{p}$ which in turn happens only if $p\cdot f(0)\equiv 0\pmod{p^2}$. But $p\cdot f(0)=a$ and therefore $a\not\equiv 0\pmod{p^2}$, since $p||a$. Thus $p\nmid [\mathcal{O}_K : \mathbb{Z}[\alpha]]$.
Of course one could use the Theorem of Dedekind which we used for the case when $p|n$ also in the case when $p\nmid n$.
A couple of references for this sort of stuff (besides those already mentioned) are a preprint of a paper by Alden Gassert which can be found here and chapters 7 and 10 of Alaca and Williams Introductory Algebraic Number Theory.